Assessment of Capacity Changes Due to Automated ...Standard Title Page - Report on Federally Funded Project 1. Report No.: 2. Government Accession No.: 3. Recipient’s Catalog No.:

Assessment of Capacity Changes Due to Automated Vehicles on Interstate Corridors http://www.virginiadot.org/vtrc/main/online_reports/pdf/21-r1.pdf

KEVIN HEASLIP, Ph.D., P.E. Associate Professor Department of Civil and Environmental Engineering Virginia Tech NOAH GOODALL, Ph.D., P.E. Senior Research Scientist Virginia Transportation Research Council BUMSIK KIM Graduate Research Assistant Department of Civil and Environmental Engineering Virginia Tech MIRLA ABI AAD Graduate Research Assistant Department of Civil and Environmental Engineering Virginia Tech

Final Report VTRC 21-R1

Standard Title Page - Report on Federally Funded Project

1. Report No.: 2. Government Accession No.: 3. Recipient’s Catalog No.:

FHWA/VTRC 21-R1

4. Title and Subtitle: 5. Report Date:

Assessment of Capacity Changes Due to Automated Vehicles on Interstate Corridors July 2020

6. Performing Organization Code:

7. Author(s):

Kevin Heaslip, Ph.D., P.E., Noah Goodall, Ph.D., P.E., Bumsik Kim, and

Mirla Abi Aad

8. Performing Organization Report No.:

VTRC 21-R1

9. Performing Organization and Address:

Virginia Transportation Research Council

530 Edgemont Road

Charlottesville, VA 22903

10. Work Unit No. (TRAIS):

11. Contract or Grant No.:

112013

12. Sponsoring Agencies’ Name and Address: 13. Type of Report and Period Covered:

Virginia Department of Transportation

1401 E. Broad Street

Richmond, VA 23219

Federal Highway Administration

400 North 8th Street, Room 750

Richmond, VA 23219-4825

Final

14. Sponsoring Agency Code:

15. Supplementary Notes:

This is an SPR-B report.

16. Abstract:

This study was designed to assess capacity changes due to the introduction of connected vehicles (CVs) and automated

vehicles (AVs) on Virginia freeway corridors. Overall, three vehicle types, including legacy vehicles (LVs); vehicles equipped with adaptive cruise control (ACC) (AVs); and vehicles equipped with cooperative adaptive cruise control (CACC) (connected

automated vehicles [CAVs]), were considered in mixed traffic scenarios. Each scenario included light-duty passenger vehicles

and heavy vehicles (HVs) with AV and CAV capabilities to determine their overall effect on capacity.

The team developed an AV and CAV driving behavior model and evaluated it on a test network. According to the testing

results, the 100% AV and 100% CAV scenarios increased road capacity by 28% and 92% over the 100% LV scenario,

respectively, on a basic freeway segment with intermediate vehicle behavior. Moreover, in the case of the HV scenario, AVs and

CAVs showed a substantial capacity increase. Simulations were also conducted on models of I-95 in Virginia, where AVs and

CAVs improved capacity compared to LVs. However, in some scenarios during congested conditions, AVs performed worse

than LVs with reduced speeds and increased travel times because of the frequent stop-and-go conditions because of short

headways. This issue was mitigated with the implementation of CAVs because of their ability to communicate and increase

string stability. Under uncongested conditions, AVs and CAVs improved throughput and reduced delays as compared to LVs but caused a small decrease in speeds and an increase in travel times. Additional simulations were performed on models of I-81 to

test the effects of extended grades and high percentages of HVs, where AVs and CAVs were found to have a high potential of

improving operations when compared to LVs. The presence of steep grades negatively affected the performance of all types of

vehicles, especially HVs, when compared to flat terrain. CAVs with their communication capabilities, particularly at high market

penetrations, were capable of achieving capacity increases over AV and LV scenarios in the selected I-81 segment.

AVs and CAVs proved capable of improving highway operations. Even in the presence of high percentages of HVs and

steep grades, vehicles equipped with AV and CAV technologies provided better performance than LVs. Ultimately, AVs and

CAVs need full market penetration to operate at their maximum potential. However, these technologies, even in mixed traffic,

could still offer operational benefits at lower penetrations.

The Virginia Department of Transportation’s (VDOT) Traffic Engineering Division should stay updated on developments in

AVs to ensure that VDOT simulation models reflect the existing and anticipated vehicle fleet. They should consider using the

capacities described in this report as guidance when calibrating models of CVs and AVs in simulations of freeway corridors. Because capacity estimates depend on AV and CAV market penetration, VDOT and the Virginia Transportation Research

Council should investigate methods to estimate the prevalence, capabilities, and rate of usage of CV and AV driving technologies

on Virginia roads.

17 Key Words: 18. Distribution Statement:

Automated vehicles, connected vehicles, adaptive cruise

control, cooperative adaptive cruise control

No restrictions. This document is available to the public

through NTIS, Springfield, VA 22161.

19. Security Classif. (of this report): 20. Security Classif. (of this page): 21. No. of Pages: 22. Price:

Unclassified Unclassified 83

Form DOT F 1700.7 (8-72) Reproduction of completed page authorized

FINAL REPORT

ASSESSMENT OF CAPACITY CHANGES DUE TO AUTOMATED VEHICLES

ON INTERSTATE CORRIDORS

Kevin Heaslip, Ph.D., P.E.

Associate Professor

Department of Civil and Environmental Engineering

Virginia Tech

Noah Goodall, Ph.D., P.E.

Senior Research Scientist


Bumsik Kim

Graduate Research Assistant


Virginia Tech

Mirla Abi Aad



Virginia Tech

In Cooperation with the U.S. Department of Transportation

Federal Highway Administration


(A partnership of the Virginia Department of Transportation

And the University of Virginia since 1948)

Charlottesville, Virginia

July 2020

VTRC 21-R1

ii

DISCLAIMER

The contents of this report reflect the views of the authors, who are responsible for the facts and the

accuracy of the data presented herein. The contents do not necessarily reflect the official views or policies

of the Virginia Department of Transportation, the Commonwealth Transportation Board, or the Federal

Highway Administration. This report does not constitute a standard, specification, or regulation. Any

inclusion of manufacturer names, trade names, or trademarks is for identification purposes only and is not to

be considered an endorsement.

Copyright 2020 by the Commonwealth of Virginia.

All rights reserved.

iii

ABSTRACT

This study was designed to assess capacity changes due to the introduction of connected

vehicles (CVs) and automated vehicles (AVs) on Virginia freeway corridors. Overall, three

vehicle types, including legacy vehicles (LVs); vehicles equipped with adaptive cruise control

(ACC) (AVs); and vehicles equipped with cooperative adaptive cruise control (CACC)

(connected automated vehicles [CAVs]), were considered in mixed traffic scenarios. Each

scenario included light-duty passenger vehicles and heavy vehicles (HVs) with AV and CAV

capabilities to determine their overall effect on capacity.

The team developed an AV and CAV driving behavior model and evaluated it on a test

network. According to the testing results, the 100% AV and 100% CAV scenarios increased

road capacity by 28% and 92% over the 100% LV scenario, respectively, on a basic freeway

segment with intermediate vehicle behavior. Moreover, in the case of the HV scenario, AVs and

CAVs showed a substantial capacity increase. Simulations were also conducted on models of I-

95 in Virginia, where AVs and CAVs improved capacity compared to LVs. However, in some

scenarios during congested conditions, AVs performed worse than LVs with reduced speeds and

increased travel times because of the frequent stop-and-go conditions because of short headways.

This issue was mitigated with the implementation of CAVs because of their ability to

communicate and increase string stability. Under uncongested conditions, AVs and CAVs

improved throughput and reduced delays as compared to LVs but caused a small decrease in

speeds and an increase in travel times. Additional simulations were performed on models of I-81

to test the effects of extended grades and high percentages of HVs, where AVs and CAVs were

found to have a high potential of improving operations when compared to LVs. The presence of

steep grades negatively affected the performance of all types of vehicles, especially HVs, when

compared to flat terrain. CAVs with their communication capabilities, particularly at high

market penetrations, were capable of achieving capacity increases over AV and LV scenarios in

the selected I-81 segment.

AVs and CAVs proved capable of improving highway operations. Even in the presence

of high percentages of HVs and steep grades, vehicles equipped with AV and CAV technologies

provided better performance than LVs. Ultimately, AVs and CAVs need full market penetration

to operate at their maximum potential. However, these technologies, even in mixed traffic, could

still offer operational benefits at lower penetrations.

The Virginia Department of Transportation’s (VDOT) Traffic Engineering Division

should stay updated on developments in AVs to ensure that VDOT simulation models reflect the

existing and anticipated vehicle fleet. They should consider using the capacities described in this

report as guidance when calibrating models of CVs and AVs in simulations of freeway corridors.

Because capacity estimates depend on AV and CAV market penetration, VDOT and the Virginia

Transportation Research Council should investigate methods to estimate the prevalence,

capabilities, and rate of usage of CV and AV driving technologies on Virginia roads.

1

FINAL REPORT

ASSESSMENT OF CAPACITY CHANGES DUE TO AUTOMATED VEHICLES

ON INTERSTATE CORRIDORS

Kevin Heaslip, Ph.D., P.E.

Associate Professor


Virginia Tech

Noah Goodall, Ph.D., P.E.

Senior Research Scientist


Bumsik Kim



Virginia Tech

Mirla Abi Aad



Virginia Tech

INTRODUCTION

In Calendar Year 2018-2021 Virginia Department of Transportation (VDOT) Business

Plan (VDOT, 2017), one of the stated agency goals was to “Ensure Efficient Highway

Operations.” One of the most challenging issues in meeting that goal is congestion on interstate

corridors. VDOT has been very proactive in the use of technology to make Virginia’s interstates

more efficient and reliable for the traveling public. Recent technology implementations by

VDOT include variable corridor pricing, advanced traveler information, and research on

connected vehicles (CVs) and automated vehicles (AVs). Much of the ongoing research on CVs

and AVs has focused on the safety considerations in the deployment of these vehicles, and there

is a need to understand the operational impacts on areas where there is currently congestion in

Virginia. The introduction of connectivity and communication in transportation infrastructure

and vehicles is expected to increase roadway capacity as vehicles can comfortably and safely

travel at short headways. Many new vehicles have technology and safety features that employ

adaptive cruise control (ACC) and lane-keeping assistance. Future AVs are likely to employ

these technologies along with connectivity to allow for increased capacity.

2

Traditional capacity estimation tools and traffic simulation software rely on models of

human driving behavior and have not been calibrated to model the complexities of the transition

to computer-driven CVs and AVs, especially not a future where CVs and AVs are the only types

of vehicles on Virginia interstates. New models are needed that capture the behavior of AVs and

CAVs in traffic to improve the accuracy of capacity estimation tools and simulation software.

In order to gain insights on the capacity impacts of AVs, simulation is needed as

cooperative adaptive cruise control (CACC) is not widely available and there is no way to

determine which drivers in the field are using ACC features at a given moment. Simulation will

allow for the ability to understand the impacts of automation of the driving process, with vehicles

adjusting to shorter following distances, thus potentially increasing capacity and reducing

variability in speeds and vehicle behavior. The adjustments to capacity may change operations

and mitigate the need for the expansion of roadways and other transportation demand

management programs. This study had the goal of providing guidance on the impact of CVs and

AVs on capacity and to help VDOT understand what levels of market penetration (MP) of AVs

will necessitate changes to design guidelines and planning practices.

PURPOSE AND SCOPE

The purpose of the study was to (1) investigate potential legal implications of AVs on

Virginia freeways; (2) quantify the change in interstate capacity due to varying levels of

penetration of automated technologies; (3) gain an understanding of the effect of automated

driving on interstate speed and throughput; and (4) assess the impacts for future VDOT planning

and operations of interstates.

The scope of the study was limited to interstate highways. The effects of technology in

both rural and urban settings were examined. The research question examined was: “What is the

effect of the implementation of AV and CAV technologies on interstate capacity in Virginia,

including consideration of MP of the technology and roadway characteristics?” This research

question led to the answering of the following operational policy question: “What actions should

VDOT take to prepare for AVs because of changes in highway capacity?” To answer these

questions, the research team investigated capacity changes using PTV VISSIM traffic simulation

software (hereinafter “VISSIM”).

A comprehensive experimental design implemented on three test geometries was the

basis for the investigation detailed in this report. The results from each of the three scenarios

illustrated different components of capacity to provide decision makers with the information

needed to make informed decisions on future transportation infrastructure investments.

3

Assumptions

The terms automated vehicles, autonomous vehicles, and connected automated vehicles

are sometimes used interchangeably in the literature, but they describe different aspects of

operation and communication. For this study, pertinent terms were defined as follows:

Legacy vehicles (LVs) are vehicles that do not have ACC or lane-keeping assistance.

These vehicles require humans to sense the environment around the vehicle and

execute the driving task without vehicle assistance.

Automated vehicles (AVs) automate some or all elements of the driving process.

Technologies that are currently automating the driving process but have not entirely

replaced the driver include basic systems such as cruise control and traction control,

more advanced systems such as ACC and lane-keeping assistance, and combinations

of these systems. For the purposes of this study, AVs refer to vehicles with ACC or

some other automated control of throttle that can respond to changing traffic

conditions.

Autonomous vehicles are fully self-driving vehicles where there is no input from the

driver in the driving task. These vehicles are a subset of AVs, which include vehicles

that automate only an individual driving task such as lane-keeping. Some

autonomous vehicles employ a safety driver that can disengage the autonomous

mode. These vehicles do not connect to other vehicles or infrastructure. Fully

autonomous vehicle technology is still in development, although industry leaders

have been active in the research and testing of autonomous vehicles.

A connected vehicle (CV) or connected automated vehicle (CAV) is a vehicle that is

capable of communicating with other vehicles or roadside infrastructure using

wireless technology. CVs that are not automated rely on the driver to conduct the

driving task, where the CAV does not require human interaction for some or all tasks.

In this study, the LV was assumed to be a vehicle without connectivity or advanced

automation technology, such as ACC and lane-keeping. AVs are considered to be completely

automated vehicles without connectivity. For CAVs, the assumption for automation is the same

as for AVs with connectivity to the roadside and other vehicles.

Study Limitations

Although this study was a comprehensive simulation study of the potential outcomes that

could occur due to the implementation of AVs on Virginia interstates, several issues cannot be

simulated easily:

4

Non-recurring congestion. The introduction of AVs promises to reduce the number

of crashes significantly and mitigate some of the effects of weather-related non-

recurring congestion. Even in areas of recurring congestion, such as I-66 in Northern

Virginia, non-recurring congestion due to crashes and police activity extends the

length of congestion and reduces the capacity of the roadway during lane blockages

or rubbernecking. Benefits tied to the reduction in non-recurring congestion were not

explicitly examined in this study.

Narrowing of lanes due to the precise latitudinal positioning of the vehicle. There

may be opportunities to narrow lane widths under high AV penetrations, potentially

allowing the creation of additional lanes within the existing right of way. If the travel

lanes are narrowed, then lane changing might be more limited by the lane

configurations; however, with limited right of way for most urban interstates, this

might be a significant development. This study does not consider lane-narrowing,

which may be possible with automated driving systems.

Exclusive lanes. This study looked only at mixed AV and LV flow in shared lanes

and did not explicitly examine exclusive AV managed lanes adjacent to mixed lanes.

The results from the basic network at 100% AV and 100% CAV provide insight on

how managed lanes would operate and the amount of capacity expected from those

lanes —restricting certain lanes to AVs could also change the capacity of corridor

level simulations for the mixed flow case. More information on dedicated lanes for

AVs can be found in National Highway Cooperative Research Program (NCHRP)

Report 891 (Booz Allen Hamilton, WSP, and New Jersey Institute of Technology,

2018).

Vehicle ownership rates. Several prominent automobile original equipment

manufacturers (OEMs) have been exploring business models of shared AV fleets

instead of private vehicle ownership. A fleet or on-demand rental model could

encourage more car sharing, which could decrease demand to levels below the

available capacity in many parts of Virginia. Alternative vehicle use and ownership

models were not explored in this study.

Trip pricing. Governments, insurers, or AV fleet operators may tax or charge per

vehicle-trip based on the anticipated congestion of the travel route at the time of the

trip. This dynamic pricing may have the effect of disincentivizing peak hour trips and

encouraging trip time spreading, which may reduce peak-period demand but increase

off-peak demand. This was not examined in this study.

Assumptions for vehicle movement. The assumptions for the vehicle movement

algorithms are detailed in this report; however, algorithms for vehicle movement in

specialty situations, such as weaving areas, were not implemented in the simulation.

5

However, merging and diverging behavior was accounted for at interchanges and

during platooning maneuvers.

Limitations of simulation. Simulation is limited by the modeling of driver behavior

for LVs and the algorithms implemented for AVs and CAVs. Current efforts by

automobile OEMs have been investigating the use of machine learning and artificial

intelligence techniques for the movement of vehicles. The algorithms implemented

for this study were derived from a state-of-the-art literature review.

LITERATURE REVIEW

Before the methodology of this study is discussed, the current legal and regulatory

environment for the behavior of automated vehicles operating on freeways in Virginia needs to

be reviewed. Likewise, past attempts to assess the impacts of CVs and AVs on capacity need to

be reviewed to assist in the development of the methodology used in this study.

Legal and Regulatory Review

The purpose of this analysis was to ascertain possible bounds on permissible behavior for

AVs both currently and in the near future as they might affect capacity modeling. Five aspects

were considered: following distances, lateral movements, lane restrictions, platooning, and

dedicating lanes for AVs. This analysis was not performed by an attorney and does not represent

any official legal interpretation by the Commonwealth of Virginia or any of its agencies. The

Office of the Attorney General has not yet issued any interpretations of many of these issues.

The list of legal and regulatory issues identified in this study is neither comprehensive nor

exhaustive. When case law is cited, it is based on the analysis of other cited legal scholarship

and may be from non-Virginia courts; these decisions may have persuasive but not binding

authority on cases in Virginia courts.

Following Distance Standards

The vehicle behavior with the most significant impact on freeway capacity is the distance

or time headway at which a vehicle follows the vehicle directly ahead. The spacing between

vehicles can be expressed in units of distance or time. Spacing can be converted between

distance and time using Equation 1:

𝑑 = 𝑣𝑡 (𝐸𝑞. 1) In this equation, d is the distance (feet or meters), t is the time (seconds), and v is the speed of the

vehicle (feet per second or meters per second).

6

In addition, there are two standard definitions of vehicle spacing: gap and headway. Gap

refers to the separation between the rear of the lead vehicle and the front of the following

vehicle. Headway refers to the separation between the front of the lead vehicle and the front of

the following vehicle.

The gap can be converted to headway using Equations 2 and 3:

𝑑ℎ𝑒𝑎𝑑𝑤𝑎𝑦 = 𝑑𝑔𝑎𝑝 + 𝑥𝑙 (𝐸𝑞. 2)

𝑡ℎ𝑒𝑎𝑑𝑤𝑎𝑦 = 𝑡𝑔𝑎𝑝 +

𝑥𝑙

𝑣(𝐸𝑞. 3)

where

𝑑ℎ𝑒𝑎𝑑𝑤𝑎𝑦 is the distance headway (feet or meters)

𝑑𝑔𝑎𝑝 is the distance gap (feet or meters)

𝑡ℎ𝑒𝑎𝑑𝑤𝑎𝑦 is the time headway (seconds)

𝑡𝑔𝑎𝑝 is the time gap (seconds)

𝑥𝑙 is the length of the lead vehicle (feet or meters)

𝑣 is the vehicle speed (feet per second or meters per second).

Driver’s Education Training

Many state departments of motor vehicles (DMVs), in their driving training courses,

recommend headways of 2 to 4 seconds (Le Vine et al., 2017). The Virginia DMV (2018)

recommends a varying headway of between 2 and 4 seconds depending on the speed of the

vehicle, as shown in Table 1.

These headways are directly related to capacity: 2-second headways yield 1,800

veh/ln/hr; 3-second headways yield 1,200 veh/ln/hr; and 4-second headways yield 900 veh/ln/hr.

Observed and theoretical capacities are higher than those suggested by the Virginia DMV

following distances. The Highway Capacity Manual (HCM) assumes a theoretical maximum

capacity of a freeway under ideal conditions of 2,250 to 2,400 passenger cars per hour per lane

depending on free-flow speed (Transportation Research Board, 2016). These equate to

headways of 1.5 to 1.6 seconds.

AVs may also follow at shorter than recommended time headways. The ACC feature in a

2017 Audi Q7 can be set to the following time gap of 1 second (Audi AG, 2016), and Virginia

recently hosted a truck platoon demonstration with gaps of 0.6 seconds at speeds of 55 mph

(Reiskin, 2017). In research demonstrations, AVs have driven with gaps as short as 13 ft and

0.18 seconds when using vehicle-to-vehicle communications (Tsugawa, 2013).

7

Table 1. Recommended Time Headways in Dry Conditions

Recommended Headway Vehicle Speed

2 seconds Under 35 mph

3 seconds 35-45 mph

4 seconds 46-70 mph

Source: Virginia Department of Motor Vehicles (2018).

Statutory Guidance

The Virginia Driver’s Manual is an informational tool. Therefore, these

recommendations do not supersede “the Code of Virginia, Virginia Administrative Code, or any

other statute” (Virginia DMV, 2018). The Code of Virginia is less specific than DMV guidance

regarding following distance. Va. Code § 46.2-816 states: “The driver of a motor vehicle shall

not follow another vehicle, trailer, or semitrailer more closely than is reasonable and prudent,

having due regard to the speed of both vehicles and the traffic on, and conditions of, the highway

at the time.” Violation of this restriction can constitute negligence per se (Smith, 2013), and the

overall effect of this law on capacity depends on the interpretation of the qualitative terms

“reasonable and prudent.”

Assured Clear Distance Ahead (ACDA) Doctrine

One attempt to define a reasonable following distance is the ACDA doctrine. This refers

to a common law manifestation of defensive driving concepts where a vehicle operator is

responsible for maintaining a sufficient gap so that the vehicle can be stopped for an object

within his or her path. Pearson (2005) defines ACDA as requiring a driver to “regulate his speed

so that he can stop within the range of his vision.” Most jurisdictions follow this standard,

presuming negligence on the part of the following driver in a rear-end collision regardless of

whether the leading vehicle was moving or stopped (Buchwalter et al., 1963).

An exception to the ACDA doctrine is the Sudden Emergency doctrine, which excuses a

driver from negligence in a collision with a vehicle, person, or object that moves unexpectedly

into the lane, either laterally or vertically, e.g., a falling tree branch. Buchwalter et al. (1963)

provides an example: “A motorist driving at a reasonable speed and obeying the rules of the road

is generally not liable for injuries to a child who darts in front of the vehicle so suddenly that the

motorist cannot avoid injuring the child, as where a child darts out from behind other vehicles

that were stopped in traffic, directly into the path of the vehicle, and there is no evidence that

[the] driver was driving too fast.” A vehicle abruptly changing lanes into the path of another

vehicle may also qualify under the Sudden Emergency doctrine (Decker v. Wofford, 1960),

whereas physical features of the roadway may not (Coppola v. Jameson, 1972).

Human drivers routinely violate the ACDA standard. In low-light conditions, drivers

adhering to ACDA would be limited to speeds of 20 mph to avoid striking a “dark-clad

pedestrian” (Leibowitz et al., 1998). Le Vine et al. (2017) note that ACDA generally requires

drivers to avoid striking both the vehicle ahead and stationary objects. Avoiding only the vehicle

8

ahead is referred to as the “weak” interpretation and allows closer following distances. Also

avoiding stationary objects is referred to as the “strong” interpretation, as it requires that a driver

maintain sufficient space to stop not only for the vehicle ahead but also for debris that may

appear within the driver’s range of vision only after the leading vehicle has passed over it. For

example, in the case in which a tractor-trailer may drive over a moderately sized box with which

a smaller following car might collide; the driver of the following vehicle would need to maintain

a following distance adequate to come to a complete stop. Disregarding reaction time, a vehicle

traveling at 55 mph and capable of decelerating at a rate of 16.4 ft/s2 (Le Vine et al., 2017) must

maintain a time gap of 2.46 seconds to the leading vehicle to avoid striking run-over debris, a

distance greater than the minimum settings on some ACC systems (Audi AG, 2016), as well as

following distances observed in the field.

Le Vine et al. (2017) calculate minimum allowable headway as shown in Equation 4:

𝐻𝑚𝑖𝑛 = 𝑡𝑙𝑎𝑔 𝑓 +𝑣

2𝑎𝑓−

𝑥𝑣𝑒ℎ −𝑣2

2𝑎𝑙

𝑣(𝐸𝑞. 4)

where

Hmin is the minimum allowable time headway (seconds)

tlag f is the reaction time of following vehicle f, set as 0.4 seconds unless otherwise noted

af is the maximum deceleration of following vehicle f (ft/s2)

al is the assumed maximum deceleration of leading vehicle l (ft/s2)

v is the free-flow speed (feet per second)

xveh is the length (units of distance) of leading vehicle l, set to 19 ft unless otherwise

noted.

The derivation of the minimum headway equation can be found in Appendix B of Le Vine et al.

(2016).

Based on a vehicle’s braking ability, the applied interpretation of ACDA, assumptions

regarding the leading vehicle’s braking ability, and vehicle speeds, there are several ways that

adherence to ACDA may impact minimum following headways and freeway capacity. Le Vine

et al. (2017) identifies 11 scenarios, shown in Table 2.

In many scenarios, the following vehicle is expected to brake at 16.4 ft/s2, which

represents both the upper limit of ACC braking standards and the lower limit of forward collision

mitigation system standards. The leading vehicle is expected to brake at various, often higher

rates of deceleration, forcing the following vehicle to leave adequate spacing. Headways varied

from 0.6 seconds for following vehicles assumed to be capable of braking at the same rate as the

lead vehicle, to 29.2 seconds for the following vehicle that can decelerate no faster than is

permitted on high-speed rail. In most scenarios, however, time headways at 75 mph range from

0.9 to 2.6 seconds.

9

Table 2. Automated Vehicle-Following Scenarios With Different ACDA Interpretations and Assumptions

Name

al

(ft/s2)

af

(ft/s2)

Min. Allowable Headway (s)

Notes 55 mph 65 mph 75 mph

Baseline

Weak

28.3 16.4 1.7 1.8 2.0 Must stop for vehicle ahead only

Baseline

Strong

Inf 28.3 2.1 2.3 2.5 Must also stop for debris

Scenario 1 21.3 16.4 1.2 1.3 1.3 Wet pavement

Scenario 2 41.6 16.4 2.1 2.4 2.6 Assumes lead vehicle is high performance

Scenario 3 N/A al 0.6 0.6 0.6 Assumes can brake at same rate as lead vehicle

Scenario 4 41.6 28.3 1.1 1.1 1.2 Following vehicle can brake at maximum rate

Scenario 5 30.38 26.21 0.8 0.8 0.9 al and af brake at 99.9th and 0.1th percentile of

typical passenger car hard brake, respectively

Scenario 6 28.3 1.8 21.6 25.4 29.2 af same as high-speed rail

Scenario 7 28.3 26.0 0.8 0.7 0.7 af set so that local maximum capacity attained at

v = 75 mph

Scenario 8 28.3 16.4 1.3 1.4 1.6 Instant reaction time using CV technologies (0.4

s all other scenarios)

Scenario 9 28.3 16.4 1.7 1.8 2.0 25% longer lead vehicle length

Source: Le Vine et al. (2017).

ACDA = Assured Clear Distance Ahead; al = leading vehicle; af = following vehicle; Inf = infinite; N/A = not

applicable; CV = connected vehicle.

Human drivers have been observed to violate ACDA requirements routinely. Video of

vehicles on a California freeway was collected and analyzed as part of the Next Generation

Simulation (NGSIM) study (Federal Highway Administration, 2007). Assuming that following

vehicles could brake at maximum rates with negligible reaction time, drivers violated ACDA

0.2% of the time (Le Vine et al., 2017). Under more realistic assumptions of reaction times

between 0.5 and 1.75 seconds, vehicles were observed to violate ACDA requirements between

1.5% and 49% of the time (Le Vine et al., 2017).

Industry Standards

ACC systems on production vehicles generally adhere to ISO 15622 standard

(International Organization for Standardization, 2010). Under this standard, vehicles are limited

to the following time gap of 0.8 seconds or higher. At least one available time gap setting should

be within the range of 1.5 to 2.2 seconds. When the ACC system is initiated and has not retained

the previous time gap setting selected by the driver, then the time gap must be set to a

“predefined default value equal of 1.5 s or greater” (International Organization for

Standardization, 2010).

Summary of Standards

Table 3 lists several of the time headways discussed in this section. All represent

vehicles with various levels of automation except the DMV headway. All but one of the Le Vine

et al. (2017) settings (Scenario 8) represent unconnected vehicles that cannot communicate

wirelessly with the lead vehicle but instead must rely on their sensors to detect closing speed.

10

Table 3. Examples of Car-Following Headways

Citation

Headway at 65

mph (s)

Notesa

Le Vine et al. 2017 0.6-2.6 Requires adherence to Assured Clear Distance Ahead

doctrine. See citation and Table 10 for parameters and

assumptions.

International Organization for

Standardization 2010

1.0 Minimum allowable headway for the ACC system.

1.7 Minimum default headway for the ACC system.

2.4 Maximum default headway for the ACC system.

Audi AG 2016 1.2 Minimum headway setting for 2017 Audi Q7’s ACC system.

2 Recommended headway setting for 2017 Audi Q7’s ACC.

Virginia Department of Motor

Vehicles 2018

4 Recommended headway, not specified in the Code of

Virginia.

ACC = adaptive cruise control. a Time gaps have been converted to headways by adding 0.2 seconds, i.e., the time required for a vehicle to travel 19

ft (the length of a vehicle) at 65 mph.

Legal Definition of Driver

A 2018 NCHRP legal audit of state motor vehicle codes for AVs noted that following

distance laws in most states, including Virginia (§ 46.2-816), refer to the “driver” of a motor

vehicle. If the term “driver” is interpreted as referring only to a natural person, then following

distance restrictions may not apply to AVs:

Following distance requirements generally apply to the “driver” of a vehicle. But recall that the

term “driver” is ambiguous and can thus have a range of meanings, which includes the possibility

that trucks with an ADS properly engaged have no “driver” and hence are not bound by any

following distance requirement at all (National Academies of Sciences, Engineering, and

Medicine, 2018).

The authors of the study recommend that policymakers consider the possibility that their

existing vehicle codes might be interpreted to regulate only “drivers” who are human, exempting

highly automated vehicles (where automated driving systems are effectively the “drivers”) from

most regulation.

Lateral Movement Statutes

The Virginia DMV recommends—but does not require—driving in the middle of the

lane, especially when driving through work zones (Virginia DMV, 2018). In addition, several

statutes address lane selection on freeways. Table 4 is a non-comprehensive list of statutes with

potential relevancy for AV operations on freeways.

The Virginia statutes listed in Table 4 may affect capacity models in several ways.

Section 46.2-842.1 of the Code of Virginia requires drivers traveling “to the left and abreast of

another motor vehicle on a divided highway” to move to the right as soon as they can safely do

so when signaled either by “audible or light signal” by an overtaking driver.

11

Table 4. Sample of Virginia Statutes Regarding Lane Selection Relevant to Automated Vehicles on Freeways

Code of Virginia Relevant Language

§ 46.2-838. Passing when

overtaking a vehicle.

The driver of any vehicle overtaking another vehicle proceeding in the

same direction shall pass at least two feet to the left of the overtaken

vehicle and shall not again drive to the right side of the highway until

safely clear of such overtaken vehicle, except as otherwise provided in this

article.

§ 46.2-804. Special regulations

applicable on highways laned for

traffic; penalty.

Any vehicle proceeding at less than the normal speed of traffic at the time

and place and under the conditions existing shall be driven in the lane

nearest the right edge or right curb of the highway when such lane is

available for travel except when overtaking and passing another vehicle or

in preparation for a left turn or where right lanes are reserved for slow-

moving traffic as permitted in this section;

§ 46.2-842.1. Drivers to give way

to certain overtaking vehicles on

divided highways.

It shall be unlawful to fail to give way to overtaking traffic when driving a

motor vehicle to the left and abreast of another motor vehicle on a divided

highway. On audible or light signal, the driver of the overtaken vehicle

shall move to the right to allow the overtaking vehicle to pass as soon as

the overtaken vehicle can safely do so. A violation of this section shall not

be construed as negligence per se in any civil action.

§ 46.2-921.1. Drivers to yield

right-of-way or reduce speed when

approaching stationary emergency

vehicles or public utility vehicles

on highways; penalties.

The driver of any motor vehicle, upon approaching a stationary vehicle

that is displaying a flashing, blinking, or alternating blue, red, or amber

light or lights as provided in § 46.2-1022, 46.2-1023, or 46.2-1024,

subdivision A 1 or 2 of § 46.2-1025, or subsection B of § 46.2-1026 shall

(i) on a highway having at least four lanes, at least two of which are

intended for traffic proceeding as the approaching vehicle, proceed with

caution and, if reasonable, with due regard for safety and traffic

conditions, yield the right-of-way by making a lane change into a lane not

adjacent to the stationary vehicle or (ii) if changing lanes would be

unreasonable or unsafe, proceed with due caution and maintain a safe

speed for highway conditions.

The provisions of this section shall not apply in highway work zones as

defined in § 46.2-878.1.

Modelers using microscopic simulation may wish to incorporate this behavior into lane-

changing models, especially if AV developers choose to implement audible or light signals when

a driver is overtaking in order to leverage this statute.

Capacity may decrease if AV drivers strictly adhere to § 46.2-921.1. Often referred to as

the “move over law,” this statute requires drivers encountering stationary vehicles with activated

blue, red, or amber lights “on a highway having at least four lanes, at least two of which are

intended for traffic proceeding as the approaching vehicle” to move into the lane not adjacent to

the stationary vehicle. If changing lanes is “unreasonable or unsafe,” the driver should “proceed

with due caution and maintain a safe speed for highway conditions.” This may decrease capacity

as vehicles switch lanes. This may also decrease capacity if AV developers determine that the

“safe speed for highway conditions” is significantly lower than free-flow speed.

12

Platooning Statutes

For this section, platoons refer to two or more vehicles that operate on roadways in a

close formation with automated driving technologies on some, but not necessarily all, vehicles.

To achieve small headways of 1 second or less—and the resulting reductions in wind resistance

and fuel usage—wireless connectivity among the vehicles is often used. Early deployments of

platooning in the United States are expected to involve the trucking industry, as it enables long-

distance shipment of goods with lower fuel costs, improved safety, and fewer paid drivers.

Platoons do not require high levels of automation, as the lead vehicle in the platoon may be

driven by a human who is responsible for slowing for traffic and avoiding incidents. The

remaining vehicles may merely follow the vehicle ahead, provided they can safely move to the

shoulder in the event of a hardware or software failure.

Virginia uses the reasonable and prudent standard for regulating following

distances (§ 46.2-816):

The driver of a motor vehicle shall not follow another vehicle, trailer, or semitrailer more closely

than is reasonable and prudent, having due regard to the speed of both vehicles and the traffic on,

and conditions of, the highway at the time.

According to a recent NCHRP report auditing state laws on vehicle automation, states

such as Virginia that use only the reasonable and prudent standard may not need to modify their

legislation to permit closely spaced platoons:

However, in states that currently apply Approach #1, the due care or “reasonable and prudent”-

type standard that allows for safe distances between vehicles to vary based “the speed of such

vehicles and the traffic upon and the condition of the highway” would arguably allow for safe

distances to vary based on the [connected and automated driving system] features that enable truck

platooning (National Academies of Sciences, Engineering, and Medicine, 2018).

Scribner (2018), in a report from the Competitive Enterprise Institute, concurs with the

finding in the NCHRP report but recommends adding language to § 46.2-816 explicitly

exempting CV and AV applications.

A recent audit of state motor vehicle law found that although state DMVs and

departments of transportation generally treat platoons as groups of individual trucks, there are

other terms in state statutes that might be interpreted to cover platoons:

To take one example, in the UVC and many of the state codes in our sample, there is a repeated

reference to a “combination of vehicles,” a term that is rarely defined. While there appears to be a

strong consensus that the term does not cover platoons, the law itself is somewhat ambiguous on

this point (National Academies of Sciences, Engineering, and Medicine, 2018).

In Virginia, the term “combination of vehicles” appears in 29 statutes but is not defined.

[These statutes include § 46.2-1117.1. Commercial delivery of towaway trailers; § 46.2-697.

13

Fees for vehicles not designed or used for transportation of passengers; § 46.2-1110. Height of

vehicles; damage to overhead obstruction; penalty; § 46.2-1104. Reduction of limits by

Commissioner of Highways and local authorities; penalties; § 46.2-644.01. Lien of keeper of

garage, § 46.2-1112. Length of vehicles, generally; special permits; vehicle combinations, etc.,

operating on certain highways; penalty; § 46.2-613.1. Civil penalty for violation of license,

registration, and tax requirements and vehicle size limitations; § 3.2-5812. Capacity of scales

not to be exceeded; determining gross or tare weight of vehicle or combination of vehicles; §

46.2-701. Combinations of tractor trucks and semitrailers; five-year registration of certain trailer

fleets; § 46.2-1116. Vehicles having more than one trailer, etc., attached thereto; exceptions; §

46.2-1117.1. Commercial delivery of towaway trailers; § 46.2-113. Violations of this title;

penalties; § 46.2-1083. Rear fenders, flaps, or guards required for certain motor vehicles; § 46.2-

1127. Weight limits for vehicles using interstate highways; § 46.2-1018. Marker lights on

vehicles or loads exceeding thirty-five feet; § 46.2-872. Maximum speed limits for vehicles

operating under special permits; § 46.2-100. Definitions; § 46.2-341.4. Definitions; § 46.2-

2000. Definitions; § 46.2-341.16. Vehicle classifications, restrictions, and endorsements; §

46.2-704. Prohibited operations; checking on weights; penalties; § 46.2-1138.2. Town

ordinances concerning weight limits on certain roads; § 46.2-1067. Within what distances

brakes should stop vehicle; § 46.2-1138.1. City ordinances fixing weight limits on certain roads;

§ 46.2-657. When registration by nonresident not required; § 46.2-843. Limitations on

overtaking and passing; § 46.2-870. Maximum speed limits generally; § 46.2-1231.1. Immunity

from liability for certain towing; § 46.2-1068. Emergency or parking brakes.] The authors of

the legal audit recommend that policymakers consider providing guidance or introducing new

legislation to provide a clearer definition of truck platoons. If platoons were classified under the

term “combination of vehicles,” the authors argued, truck platoons would probably violate most

length and weight restrictions (National Academies of Sciences, Engineering, and Medicine,

2018). In Virginia, vehicle lengths are covered under § 46.2-1112, which in most circumstances

limits combinations of vehicles to 65 ft. Gross weight for a combination of vehicles on interstate

highways is generally limited to 80,000 lb (84,000 lb with a fee) under § 46.2-1127, § 46.2-1126,

and § 46.2-1128.

Past Studies of CV and AV Capacity

AV Capacity on a Basic Freeway Segment

The HCM defines a basic freeway segment as “outside the influence area of any merge,

diverge, or weaving segments and of any signalized intersections” (Transportation Research

Board, 2016). The basic freeway segment does not include interactions with merging, diverging,

or weaving sections, and the capacity evaluation is dependent on free-flow speed, where a 75

mph free-flow speed correlates to an estimated base capacity of 2,400 pc/hr/ln. The capacity

analysis is used to determine the number of lanes necessary to achieve a target level of service in

design and to identify what conditions lead to capacity exceedance (Teodorović and Janić, 2017).

In a basic freeway segment, AV longitudinal assistance is anticipated to have an impact through

14

the platooning of vehicles, where short following distances and uniform maneuvering can

improve traffic flow operations and throughput.

Simulation Studies

Huang et al. (2000) used simulation to evaluate the impact of AVs on a highway. In this

study, AVs were able to select maneuvering movements from eight alternatives. Based on a

simulation with mixed LVs and AVs, the introduction of AVs at more than 70% increased the

capacity from 2,000 veh/hr/ln to approximately 5,000 veh/hr/ln. In addition, the study suggested

possible AV operational parameters such as average speed, maximum deceleration, desired

headway, and a decision tree for AV behavior. The assumptions in this study used much higher

speeds than allowed on roadways in Virginia.

Kesting et al. (2008) focused on investigating control strategies for an ACC system

capable of adapting to various traffic situations. The authors’ contributions focused on

investigating the response of an ACC system to various traffic conditions when the driver

specifies a time headway and initial velocity. The authors simulated an on-ramp bottleneck and

considered traffic conditions in that bottleneck that included upstream (before), within

(congested), and downstream (after).

Talebpour and Mahmassani (2016) also simulated LVs, CVs, and AVs with different car-

following models to investigate the impact on traffic flow stability and throughput. The authors

concluded that CVs and AVs improve string stability of traffic flow and AVs prevent shock

wave formation and propagation better than CVs. Moreover, in the simulation, the throughput at

90% AV (0% CV) was 2,500 veh/hr/ln whereas the same MP of CV 90% (0% AV) was 2,200

veh/hr/ln.

Delis et al. (2015) investigated a car-following model to evaluate traffic flow dynamics

for ACC-compatible vehicles. The critical component of the model is that the relaxation time in

the ACC system is related only to the direct leading vehicle. The results demonstrated benefits

in ACC-capable vehicles in terms of flow and capacity.

VanderWerf et al. (2001, 2002) provided a mathematical model for ACC vehicle

simulation purposes. The study provides operational parameters for ACC vehicles in which a

headway time gap of 1.4 seconds was observed. The acceleration and deceleration parameters

were set at 2 m/s2 and -3 m/s2, respectively, and the vehicles’ desired speed was set at 29 m/s (65

mph). A highway capacity of 2,200 veh/hr/ln was found for ACC-compatible vehicles at 100%

penetration.

Field Studies

The effect of AVs on basic freeway segments was investigated as early as 1996 by

researchers in the Automated Highway System Program (Raza and Ioannou, 1996). Kanaris et

15

al. (1997) completed a spacing and capacity evaluation for an automated highway system by

considering fully autonomous vehicle operation and levels of infrastructure-supported vehicle

automation. The study provided a base case consideration of operational parameters that adapted

to present-day analysis with present-day vehicle operations. In addition, the vehicle throughput

mathematical models can be referenced for capacity estimation when a sensitivity analysis of

operational parameters is performed.

Shladover et al. (2012) evaluated ACC capacity by using field data collected from drivers

in an ACC vehicle to estimate time headway spacing. In this study, 31.1% of the sampled

drivers selected 2.2 seconds of headway spacing, 18.5% of the drivers selected 1.6 seconds, and

50.4% of the drivers were comfortable selecting 1.1 seconds. The authors found little capacity

effect due to the consideration of ACC vehicles with capacity ranges of 2,030 to 2,100 veh/hr/ln

with full ACC MP. The results provided in this study give a reference to actual collected data

from drivers in terms of their spacing comfort when using ACC technology.

CAVs on a Basic Freeway Segment

Simulation Studies

The connectivity of AVs provides the ability to coordinate the movement of vehicles on

the roadway. CACC is the most critical technology for enabling CAVs. The vehicle-to-vehicle

(V2V) communication capabilities of CAVs can minimize following distances in environments

where CACC vehicles are present in high penetrations and coordinate acceleration and braking

actuation in platoons of vehicles to ensure smooth traffic flow. Smooth traffic flow is not

possible with LVs or AVs because there is no coordination between vehicles in either of those

two types of vehicles.

In the previously discussed studies of ACC capacity evaluation by VanderWerf et al.

(2001, 2002), the authors also examined capacity estimates of vehicles with CACC capabilities.

Although many of the operational parameters of the CACC vehicles were the same as with ACC,

the desired headway gap was reduced to 0.5 seconds. In platoons of CACC vehicles, the change

in the following distance resulted in capacity observations of up to 4,550 veh/hr/ln with 100%

CACC MP.

van Arem et al. (2006) modeled CACC using the MICroscopic model for Simulation of

Intelligent Cruise control (MIXIC), a car-following model, and evaluated CACC based on the

number of shock waves, average speed, and capacity. The simulation was conducted on a

highway at a bottleneck segment with and without a CACC exclusive lane. The results showed

that high CACC MP leads to a few shock waves and high average speed. In addition, the authors

concluded that 60% and 80% CACC MP significantly increase the highest maximum traffic

volume after the bottleneck.

16

Zhao and Sun (2013) evaluated capacity considering ACC vehicles in a platooned

environment for a basic freeway segment. The authors’ main contribution focused on discussing

the implementation of the Intelligent Driver Model (IDM) (a car-following model) in VISSIM to

implement AV behavior. The authors estimated the maximum capacity of nearly 3,000 veh/hr

under 100% CACC MP and 10 vehicle platoons.

Ntousakis et al. (2015) simulated ACC vehicles by using the AIMSUM simulation

software with the Gipps car-following model and IDM. In the simulation, the authors used the

Gipps model for AVs and IDM for LVs because IDM can reflect LV stop-and-go waves,

contrary to the Gipps model, which was not able to show the waves. The results showed that

high ACC MP rates result in high capacity, with a capacity of approximately 3,600 veh/hr/ln at

100% of CACC penetration time gaps set between 0.8 and 1.1 seconds.

Melson et al. (2018) investigated CACC characteristics from a planning perspective using

MIXIC and the Link Transmission Model for dynamic network assignment. Capacity

improvements due to the introduction of CACC might cause significant congestion because of

the Braess paradox; the modification of a road network to improve traffic conditions may worsen

its actual traffic condition. Both MIXIC and the Link Transmission Model showed that CACC

performs better than LVs. Also, MIXIC showed that the capacity of CACC is 4,500 veh/hr/ln

whereas the capacity of LVs is 2,300 veh/hr/ln.

Shelton et al. (2016) simulated the traffic impacts of CAVs in an urban setting. The

researchers used a multi-resolution model that combines three types of modeling: macroscopic,

mesoscopic, and microscopic. From the microscopic simulation parameter validation step,

CAVs mimic the LVs and CACC vehicles with a capacity of 2,424 veh/hr/ln and 3,952 veh/hr/ln,

with each at 100% penetration.

Field Studies

Shladover et al. (2012) evaluated ACC capacity by using field data collected from actual

drivers in a CACC vehicle to estimate time headway spacing. From the authors’ observations,

12% of drivers chose a time space setting of 1.1 seconds, 7% chose 0.9 seconds, 24% chose 0.7

seconds, and 57% chose 0.6 seconds. The authors found a significant capacity effect due to the

consideration of CACC vehicles with a capacity as high as 4,000 veh/hr/ln under assumptions of

100% CACC MP. Su et al. (2016) compared ACC field data to the IDM using ACC-equipped

vehicles. The study showed the biggest root mean square error for distance gap, with speed

differential being much more accurate.

An ACC and CACC study was conducted by Milanés and Shladover (2014). Four

Infinity M56s equipped with ACC were modified with communication equipment to have CACC

functionality. The research team found that ACC has an overshooting problem resulting in the

following vehicle’s acceleration being higher than that of the leading vehicle and that CACC

could eliminate the problem. The team suggested a new car-following model that could replace

17

IDM, as the model has an undershooting problem as well as a time gap error. The proposed

ACC/CACC model uses a gap error that is calculated by comparing the location of the subject

vehicle and preceding vehicle, desired time gap setting, and speed of the subject vehicle.

CAV Platoons

Platooning is defined as several vehicles traveling close together in a single lane. The

communication link in CAVs allows for scenarios where platoons perform coordinated

maneuvers, uniform braking, and intelligent decision-making strategies to minimize driver delay

and increase highway capacity. Significant capacity impacts on a highway setting can be

attributed to the effectiveness of forming and maintaining platoons at close distances.

Work by Fernandes and Nunes (2015) focused on developing CAV platooning

algorithms to maintain highway capacity during transitions. Their research identified four

algorithms as possible transition scenarios: platoon new leader’s positioning algorithm, platoon

vehicles’ positioning algorithm, platoon joining maneuvers management algorithm, and extra

spacing for secure maneuvering improvement algorithm. The work considers merging and

diverging scenarios and capacity estimates with the use of the Simulation for Urban Mobility

traffic simulator.

Songchitruksa et al. (2016) simulated CAV behavior by using VISSIM. They defined the

CACC platoon formation rules as follows:

First, two CACC-equipped vehicles are in the same lane with 100 m spacing.

Second, the following CACC vehicle should follow for 10 seconds to join a platoon.

Third, the following vehicles in a platoon have the desired time gap from a

multinomial distribution that has four levels (0.6, 0.7, 0.9, and 1.1 second).

The study suggested that platoon dissipation conditions commence when one of the following

occurs:

a wireless signal from the wireless reception model drops

the vehicle is the 11th vehicle in a platoon (maximum 10 vehicle platoon)

a non-CACC vehicle cuts into the platoon.

The results showed that the CACC platoon system increased freeway throughput.

Benefits were greatest when the freeway had high volumes and high CACC penetration rates and

CACC vehicles traveled in the left lane.

Different platooning scenarios have been studied in conjunction with CAVs. Kesting et

al. (2008) considered platooning in the evaluation of highway capacity and provided unique

18

combinations of possible scenarios and operational parameters of 10 and 20 vehicle platoons.

Carbaugh et al. (1998) considered platooning evaluations, where vehicles followed between 1

and 2 m. The focus of Fishelson et al. (2013) was on evaluating platooning safety in a CAV

environment; platoon sizes of 3, 5, 6, 9, 11, 13, 15, 17, 19, and 21 vehicles were considered.

VanderWerf et al. (2001) considered ACC and CACC platoons in their evaluation of capacity

with a platoon size of 20 vehicles. Last, Zhao and Sun (2013) evaluated platooning scenarios

with CACC vehicles with 6 vehicles in a platoon.

Heavy Vehicles (HVs)

Little CAV research has focused on HVs, but the impact of heavy CAVs could be

significant on many corridors, such as the I-81 corridor in Virginia. Raza and Ioannou (1996)

stated that evaluation of the impact of CAVs on commercial HVs on traffic congestion and

economic growth was necessary. Raza and Ioannou stated two reasons why commercial HVs are

likely to be early adopters of AV technology. First, the average truck travels 6 times as much

and consumes 27 times more fuel than an average passenger car. Second, the average cost of a

commercial HV is 5 times higher than that of an average passenger vehicle, making the

additional cost of automation technologies a smaller proportion of total vehicle cost. Automated

highway systems (Raza and Ioannou, 1996) provide operational parameter considerations for

trucks and HVs (Kanaris et al., 1997; Kanellakopoulos and Tomizuka, 1997). Kesting et al.

(2008) provided additional HV operational parameters and vehicle considerations. The authors

restricted the truck’s performance, such as the desired speed, safe time gap, and maximum

acceleration, in IDM to replicate the truck’s behavior. This research was conducted in the 1990s

and constituted the fundamental formulations for the next generation of AVs.

Table 5 summarizes the operational parameters found in the literature of most value to

the study methodology. Appendix A provides a synthesis of the studies in this section. These

details provide useful information not included in this section due to the emphasis on AV

operational parameters. The gaps in the research that were seen were robust studies of

penetration rates of the different technologies and studies of the effects of these technologies in

actual corridors where there are interactions caused by weaving segments and bottlenecks.

19

Table 5. Summary of Findings for Operational Parameters for LVs, AVs, and CAVs

Source

Operational Parameters (LV | ACC | CACC)

Capacity (veh/hr/ln)

LV AV CAV

Ioannou et al. 1996 Avg. speed (ft/s): (N/A | 88 | 88)

Comfortable acceleration (ft/s2): (N/A | 4.83 |

4.83)

Comfortable deceleration (ft/s2): (N/A | 3.22 |

3.22)

Desired headway (s): (N/A | 0.416 | 0.72)

3,850 5,810

Huang et al. 2000 Avg. speed (ft/s): (82.02 | 82.02 | N/A)

Max. deceleration (ft/s2): (15 | 15 | N/A)

Desired headway: (N/A | 1 s + 10 m | N/A)

2,000 4,750

VanderWerf et al.

2001

Desired headway (s): (N/A | 1.4 | 0.5) 2,050 2,200 4,550

VanderWerf et al.

2002

Avg. speed (ft/s): (95.14 | 95.14 | 95.14)

Comfortable acceleration (ft/s2): (9.65 | 9.65 |

9.65)

Comfortable deceleration (ft/s2): (9.65 | 9.65 |

9.65)

Desired headway (s): (1.1 | 1.4 | 0.5)

2,099 2,142 4,259

VanderWerf et al.

2001

Traffic mix: 20% LV, 20% ACC, 60% CACC

2,100-

2,900

Freckleton et al. 2013 Vehicle length (ft): (16.4 | N/A | 16.4)

Avg. speed (ft/s): (110 | N/A | 110)

Max. acceleration (ft/s2): (8.2 | N/A | 8.2)

Max. deceleration (ft/s2): (8.2 | N/A | 8.2)

Intra-platoon spacing (ft): (3.28 | N/A | 3.28)

Inter-platoon spacing (ft): (98.43 | N/A | 98.43)

2,257

8,601

Shladover et al. 2012 Vehicle length (ft): (15.42 | 15.42 | 15.42)

Avg. speed (ft/s): (95.33 | 95.33 | 95.33)

Comfortable acceleration (ft/s2): (6.56 | 6.56 |

6.56)

Comfortable deceleration (ft/s2): (6.56 | 6.56 |

6.56)

Desired headway (s): (1.48-1.8 | 1.1-2.2 | 0.6-1.1)

2,018 2,030-

2,100

3,970

Fernandes 2015 Vehicle length (ft): (N/A | N/A | 9.84)

Avg. speed (ft/s): (N/A | N/A | 50.03)

7,200

Shelton et al. 2016 Avg. speed (ft/s): (85.07-105.6 | N/A | 85.07-

105.6)

2,424

3,952

Melson et al. 2018 Vehicle length (ft): (14.6 | N/A | 14.6)

Avg. speed (ft/s): (73.33 | N/A | 73.33)

Desired headway (s): (N/A | N/A | 0.6)

Desired spacing (ft): (6.5 | N/A | N/A)

2,300

4,500

LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles; ACC = adaptive cruise

control; CACC = cooperative adaptive cruise control; N/A = not applicable.

METHODS

Four tasks were performed to achieve the study objectives:

20

1. custom simulation development

2. simulation scenario development

3. simulation plan development

4. analysis of test networks.

Custom Simulation Development

This section describes the development of the custom simulation for CVs and AVs. This

simulation tool was designed in order to be able to simulate the interaction of LVs, AVs, and

CAVs.

Simulation Software

VISSIM was used to assess the selected Virginia highway corridors. The research team

used VISSIM 10 with its dynamic link library (DLL) (compiled by C++) to model car-following,

lane changing, and communication. Many studies have used VISSIM’s external driving

behavior model to simulate AVs and CAVs (Melson et al., 2018; Songchitruksa et al., 2016;

Zhao and Sun, 2013). At the time of this study, PTV had not released the VISSIM CAV model;

however, the research team consulted with PTV in the development of the models for this study.

Two external driving modules have been widely used in VISSIM: the COM interface and

the external driving behavior model (Drivingbehavior.dll). The COM interface can access all

data inside VISSIM and can control some aspects of the simulation, such as vehicle insertion, but

it cannot explicitly control lateral movement. On the other hand, the external driving behavior

model receives the current state of the vehicle and its surroundings from VISSIM, and the DLL

computes the acceleration/deceleration of the vehicle as well as the lateral movement behavior.

Then, the external driving behavior model passes these values back to VISSIM to be used in the

current time step. In this study, the research team used the COM interface to run multiple

simulations and export results and used the external driving behavior model for car-following,

lane changing, CACC maneuvering, and forming a platoon. The following subsections detail the

development of the car-following algorithms inserted into VISSIM through the use of an external

DLL.

Car-Following Models for LVs, AVs, and CAVs

The research team reviewed six car-following models used widely in microscopic traffic

simulation. The Gipps car-following model has been used in many studies; however, some noted

that near the end of the bottleneck (Spyropoulou, 2007), the Gipps model shows the unrealistic

behavior named the “pinch effect.” The pinch effect is a congested state that consists of a

stationary downstream front at the on-ramp bottleneck; homogeneous, lightly congested traffic

near a ramp; and velocity oscillations (“small jams”) farther upstream (Treiber et al., 2010).

IDM (Kesting et al., 2010) also has been used widely in microscopic traffic simulation in recent

21

studies investigating AVs. This model is a collision-free and complete model, as is the Gipps

model. The Milanés and Shladover (2014) model was derived from data obtained from the tests

of four production vehicles in California. MIXIC has also been used widely for simulating ACC

and CACC (Milanés and Shladover, 2014). After a review of the car-following models, MIXIC

was chosen as the model to be implemented for this study. Table 6 summarizes the advantages

and disadvantages of each car-following model.

Table 6. Advantages and Disadvantages of Each Car-Following Model

Model Advantages Disadvantages

Gipps Collision-free

Complete model

Shows unrealistic behavior in a specific

condition

IDM/

Modified IDM Collision-free

Complete model

Many modified versions exist

Need to compile for VISSIM

Milanés and

Shladover Verified by actual test drive data Tested in specific condition (highway setting)

MIXIC Collision-free

Complete model

Straightforward

Need to compile for VISSIM

Wiedemann VISSIM’s default model Not widely used for AVs and CAVs

IDM = Intelligent Driver Model; MIXIC = MICroscopic model for Simulation of Intelligent Cruise control; AV =

automated vehicle; CAV = connected automated vehicle.

Selected Car-Following Model (MIXIC)

The research team determined that a modified MIXIC was the most appropriate model

because it was developed for intelligent vehicles and the model has been used in CAV research

similar to this study (Federal Highway Administration, 2015). Because the model is relatively

simple, it is easy to understand the vehicle’s current condition and movement within the

simulation. Modifications to the model were made for the parameters of the vehicles tested in

this study.

MIXIC was developed in 1995 by van Arem et al. (1995) to investigate the impact of

ACC on traffic flow. The model assumes that the driver tries to match the speed of the

preceding vehicle. At the same time, the driver attempts to keep the space gap at the desired

value. The model was revised in 2006 by the authors to incorporate CACC characteristics (van

Arem et al., 2006). Equations 5 and 6 represent the vehicle acceleration under MIXIC:

𝑎𝑟𝑒𝑓 = 𝑥��(𝑡 + 𝛥𝑡) = 𝑚𝑖𝑛(𝑎𝑛𝑟𝑒𝑓𝑣 , 𝑎𝑛

𝑟𝑒𝑓𝑑) (𝐸𝑞. 5)

𝑑𝑛

𝑚𝑎𝑥 ≤ ��𝑛(𝑡 + 𝛥𝑡) ≤ 𝑎𝑛𝑚𝑎𝑥 (𝐸𝑞. 6)

where

𝑎𝑟𝑒𝑓 is the acceleration needed for the vehicle to reach its desired speed (𝑚/𝑠𝑒𝑐2)

22

𝑎𝑛𝑟𝑒𝑓𝑣

is the acceleration based on the difference between the current and desired speed

(𝑚/𝑠𝑒𝑐2)

𝑎𝑛𝑟𝑒𝑓𝑑

is the acceleration based on the values of the current speed, desired speed, and

distance to the downstream vehicle (𝑚/𝑠𝑒𝑐2)

𝑎𝑛𝑚𝑎𝑥 is the maximum acceleration (𝑚/𝑠𝑒𝑐2)

𝑑𝑛𝑚𝑎𝑥 is the maximum deceleration (𝑚/𝑠𝑒𝑐2)

𝑡 is time (𝑠𝑒𝑐)

�� is the vehicle speed (𝑚/𝑠𝑒𝑐)

�� is the vehicle acceleration (𝑚/𝑠𝑒𝑐2).

The difference between current and desired speed can be calculated using a constant-

speed error factor (𝑘), desired speed, and current speed in accordance with Equation 7:

𝑎𝑛𝑟𝑒𝑓𝑣 = 𝑘(��𝑛

𝑑𝑒𝑠 − ��𝑛(𝑡)) (𝐸𝑞. 7) where

𝑎𝑛𝑟𝑒𝑓𝑣

is the difference between current and desired speed (𝑚/𝑠𝑒𝑐)

��𝑛(𝑡) is current speed (𝑚/𝑠𝑒𝑐)

��𝑛𝑑𝑒𝑠 is desired speed (𝑚/𝑠𝑒𝑐)

𝑘 is the constant-speed error factor.

The difference between the current and desired speed and distance considers three

factors: acceleration, speed, and space gap (see Eq. 8). First, the model takes the preceding

vehicle’s acceleration (��𝑛−1) into account. Second, the model takes the speed differences

between the leading vehicle and the following vehicle (��𝑛−1 − ��𝑛) into account. Third, the

model considers the gap between a reference space gap and current space gap ((𝑥𝑛−1(𝑡) −

𝑥𝑛(𝑡)) − 𝑠𝑛𝑟𝑒𝑓

).

𝑎𝑛𝑟𝑒𝑓𝑑 = 𝑘𝑎��𝑛−1(𝑡) + 𝑘𝑣(��𝑛−1(𝑡) − ��𝑛(𝑡)) + 𝑘𝑑((𝑥𝑛−1(𝑡) − 𝑥𝑛(𝑡) − 𝐿𝑛−1) − 𝑠𝑛

𝑟𝑒𝑓) (𝐸𝑞. 8)

where

𝑘𝑎, 𝑘𝑣 , 𝑘𝑑 are model parameters

𝑠𝑛𝑟𝑒𝑓

is the reference space gap between vehicle n and n-1 (𝑚)

�� is vehicle acceleration (𝑚/𝑠𝑒𝑐2)

�� is vehicle speed

𝑥 is vehicle position (𝑚).

Equations 9 through 11 determine the reference gap:

23

𝑠𝑛𝑟𝑒𝑓

= 𝑚𝑎𝑥(𝑠𝑛𝑠𝑎𝑓𝑒

, 𝑠𝑛𝑠𝑦𝑠𝑡𝑒𝑚

, 𝑠𝑛𝑚𝑖𝑛) (𝐸𝑞. 9)

𝑠𝑛𝑠𝑎𝑓𝑒

=(��𝑛(𝑡))2

2(

1

𝑑𝑛−1𝑚𝑎𝑥

−1

𝑑𝑛𝑚𝑎𝑥

) (𝐸𝑞. 10)

𝑠𝑛

𝑠𝑦𝑠𝑡𝑒𝑚= 𝑇𝑛��𝑛(𝑡) (𝐸𝑞. 11)

where

𝑠𝑛𝑠𝑎𝑓𝑒

is safe following distance (𝑚)

𝑠𝑛𝑠𝑦𝑠𝑡𝑒𝑚

is system time setting following distance (𝑚)

𝑠𝑛𝑚𝑖𝑛 is minimum allowed distance (𝑚)

𝑇𝑛 is desired time gap between vehicle n-1 and n (𝑠𝑒𝑐)

dnmax is distance spacing between vehicle and upstream vehicle (𝑚)

�� is vehicle speed (𝑚/𝑠𝑒𝑐).

The model parameters 𝑘, 𝑘𝑎, 𝑘𝑣, and 𝑘𝑑 were given from the default model description

(see Table 7). However, the parameters for CACC simulation were revised because the new

parameters show the smoother and faster reaction of the CACC controller (van Arem et al.,

2006). Accordingly, the study applied the modified model parameters for LVs, AVs, and CAVs.

Table 7. MIXIC Parameters

Parameter Default Value Value From van Arem et al. (2006)

𝑘 0.3 0.3

𝑘𝑎 (acceleration) 1.0 1.0

𝑘𝑣 (velocity) 3.0 0.58

𝑘𝑑 (distance) 0.2 0.1

Operational Parameters

Table 8 summarizes the vehicle’s operational parameters, which were informed from

previous research. To reflect adequately the diverse operating conditions of AVs and CAVs,

ranges for each parameter were considered. This assumption is important because it is likely that

different vehicle manufacturers would have different settings for different vehicles. Vehicles

will also have passenger-determined settings for comfort, much like those seen in today’s ACC

systems.

There are four levels of parameters used to reflect the different operational parameters of

the vehicles. The intermediate parameters represent the average parameters of previous research.

The conservative and aggressive parameters represent two cases where the passengers could set

the parameters based on their comfort level. A very conservative condition based on

manufacturer crash liability is considered. It is expected that the first AV will operate in this

manner unless legal frameworks are changed significantly.

24

Table 8. Operational Parameters of Passenger Vehicles

Scenario

Manufacturer Liability

Conservative

Intermediate

(Base Parameters)

Aggressive

Vehicle length (ft) 14.70

Desired speed (ft/s) 52.26** 52.26** 80.00 120.00*

Acceleration (ft/s2) Desired 2.29* 2.29* 4.93 9.08**

Max. 3.28* 3.28* 7.16 13.12*

Deceleration (ft/s2) Desired 2.01* 2.01* 6.26 14.72**

Max. 8.00 8.00 13.36 25.62**

Time gap (s) AV 2.00 1.40* 1.07 0.70*

CAV 2.00 1.16* 0.65 0.42**

Space gap (ft) AV 21.00 21.00 16.05 10.50

CAV 17.40 17.40 9.75 6.30

Standstill spacing (ft) 5.00

Communication delay (ms) 100.00

** Value from 𝜇 ± 2𝜎; * Value from min./max.

AV = automated vehicles; CAV = connected automated vehicles.

The operational parameters for the LVs are the default distribution of human driving

behavior found in VISSIM. Equations 12 and 13 were used to determine the conservative and

aggressive parameters based on what was found in past research:

𝑃𝐶𝑖 = max(𝑃𝑚𝑖𝑛

𝑖 , 𝑃𝜇𝑖 − 2𝑃𝜎

𝑖) (𝐸𝑞. 12)

𝑃𝐶𝑖 = max(𝑃𝑚𝑎𝑥

𝑖 , 𝑃𝜇𝑖 + 2𝑃𝜎

𝑖) (𝐸𝑞. 13)

where

𝑃𝐶𝑖 is Parameter i’s conservative scenario value

𝑃𝐴𝑖 is Parameter i’s aggressive scenario value

𝑃𝑚𝑖𝑛𝑖 is Parameter i’s minimum value from the literature

𝑃𝑚𝑎𝑥𝑖 is Parameter i’s maximum value from the literature

𝑃𝜇𝑖 is Parameter i’s average value from the literature

𝑃𝜎𝑖 is Parameter i’s standard deviation value from the literature.

Equations for aggressive and conservative scenarios are swapped for time and space gap,

where small values represent aggressive scenarios. In the case of the space gap, the same

methods as were used in the intermediate parameters are applied, meaning an average bounded

with the minimum and maximum distributed normally. Last, the communication delay was used

as a uniform value because of the lack of studies and simulation resolution available.

In the case of HVs (light trucks [LTs] and heavy trucks [HTs]), the research team

collected operational parameters for trucks from the previous studies and used the average values

as the intermediate case. However, since the operational parameters in the literature were

limited, this study estimated the different levels of operational parameters based on the ratio of

25

passenger vehicles’ different levels of operational parameters. The power and weight values

represent ranges in the simulation. The values for these vehicles are presented in Tables 9 and

10.

Table 9. Operational Parameters of Light Trucks

Scenario


Conservative

Intermediate

(Base Parameters)

Aggressive


Weight (lb) 7000-30000

Desired speed (ft/s) 48.99 48.99 75.00 112.50

Acceleration (ft/s2) Desired 1.22 1.22 2.62 4.83

Max. 1.75 1.75 3.81 6.98

Deceleration (ft/s2) Desired 1.07 1.07 3.33 7.83

Max. 4.26 4.26 7.11 13.63

Time gap (s) AV 3.50 2.62 2.00 1.31

CAV 3.50 2.14 1.20 0.78

Space gap (ft) AV 39.25 39.25 30.00 19.63

CAV 32.12 32.12 18.00 11.63



Power (hp) 200.00-536.00


Table 10. Operational Parameters of Heavy Trucks

Scenario


Conservative

Intermediate

(Base Parameters)

Aggressive


Weight (lb) 26000-80000

Desired speed (ft/s) 45.73 45.73 70.00 105.00

Acceleration (ft/s2) Desired 0.61 0.61 1.31 2.41

Max. 0.87 0.87 1.90 3.48

Deceleration (ft/s2) Desired 0.95 0.95 2.96 6.96

Max. 3.78 3.78 6.32 12.12

Time gap (s) AV 3.50 2.62 2.00 1.31

CAV 3.50 2.14 1.20 0.78

Space gap (ft) AV 39.25 39.25 30.00 19.63

CAV 32.12 32.12 18.00 11.63



Power (hp) 400.00-600.00


Vehicle Longitudinal Movement

Movement of LVs and AVs

The difference between LVs and AVs is that the AV has a shorter time headway. Also,

the desired speeds of LVs and AVs are different because each LV has a different desired speed

26

since human drivers have different speed preferences whereas AVs follow traffic rules, such as

the speed limit. These two parameters act differently in MIXIC.

First, the time headway affects system time setting following distance in MIXIC. If the

time headway (𝑇𝑛) is larger than the system time setting, following distance (𝑠𝑛𝑠𝑦𝑠𝑡𝑒𝑚

) is likely to

be chosen as the reference space gap (snref). If the difference between current and desired speed

and distance (anrefd) is smaller, then the model will choose that acceleration since the model

compares and chooses the minimum value between the current and desired speed and distance

(anrefd) and the difference between current and desired speed (𝑎𝑛

𝑟𝑒𝑓𝑣).

The difference between the current speed and the desired speed affects the vehicle’s

acceleration. MIXIC uses differences between desired and current speed (xndes − xn(t)) to

determine the subject vehicle’s acceleration. If all the vehicles have the same desired speed and

if the traffic has no congestion, all the vehicles drive at the desired speed with string stability.

However, if all of the vehicles have different desired speeds, some vehicles may accelerate faster

than other vehicles, but the overall speed decreases due to the shock wave.

Desired Speeds

There are different desired speeds for each of the scenarios. The research team assumed

that the AVs and CAVs would travel as fast as legally allowed, and this assumption is reflected

in the desired speed for vehicles. The desired speeds for each scenario are as follows:

Basic freeway segment: according to Figures 8, 9, and 10

I-95: the speed limit as posted (55-60 mph)

I-81: 70 mph, the posted speed limit.

CAV Movement

ACC

ACC is a technology that adjusts the vehicle speed automatically to maintain a set

distance, often expressed as a time gap, from the vehicle directly ahead of the ACC vehicle. The

driver sets the maximum speed, and then a radar sensor measures the distance from the vehicle

ahead and adjusts the acceleration of the vehicle to stay the user-set distance from the vehicle

ahead. Current ACC-equipped vehicles have between three and seven distance settings

depending on the make and model. ACC is usually paired with other driver assistance features

such as emergency braking and collision avoidance.

27

CACC

CACC is an enhancement of the ACC system with wireless communication to preceding

vehicles or infrastructure to augment the ACC sensing capability. CACC technology enables the

vehicle to reduce its headway because of the communication-enabled short reaction times. Each

platoon has three different roles: a leader (first vehicle), members, and a tail (last vehicle). The

leader of the platoon acts as an AV. The distance between platoons needs to be significantly

longer than that between vehicles in the platoon. In the model, the CAV leader is determined by

adjacent vehicle type and its distance. If the subject CAV has a follower within a certain

distance and the following vehicle is also a CAV, the subject CAV becomes the leader of the

platoon. Members of the platoon are the CAVs between the leader and the tail. The headway of

a member is shorter than that of the leader because of communication capability. The tail of the

platoon drives the same as a member, and the maximum platoon size determines the length of the

platoon. Figure 1 shows each role in a platoon.

The CACC system determines the vehicle status based on the headway and the space gap

between two consecutive CAVs. The model requires that the two CAVs are close enough to

maintain a platoon based on time or space headway. If the two consecutive CAVs are not close

enough to maintain a platoon, the following vehicle tries to join a platoon from a significant

distance. If the two consecutive CAVs are close enough to maintain a platoon, the following

vehicle tries to keep that position. Figure 2 shows the CACC maneuvering algorithm.

Figure 1. The Roles Within a CAV Platoon. CAV = connected automated vehicle; CACC = cooperative

adaptive cruise control.

28

Vehicle type=CAV?

Leader dist < CACC dist

yes

Leader Type == CAV

Yes

CACC Indicator ++ 1

yes

CACC Indicator < MaxPlatoonSize

CACC Member

Yes

noCACC Indicator == MaxPlatoonSize

CACC Tail

yes no

no

CACC LeaderNot CAV

no

no

CAV but not In CACC

Figure 2. CACC Platoon Forming Algorithm. CACC = cooperative adaptive cruise control; CAV =

connected automated vehicle; dist. = distance.

CACC Maneuvers

Members and tails of the platoon can make CACC maneuvers. The leader follows a

predefined car-following model with the same headway as an AV. The CACC joining maneuver

is made only when the joining vehicle is moving slower than the desired speed and the leading

vehicle is a platoon leader. The following CAV follows the leading vehicle to make a platoon.

AVs and CAVs are required to obey traffic laws (speed limits), whereas LVs can disobey traffic

laws. Once the vehicle is in a platoon, the subject vehicle drives with the platoon unless (1) the

subject vehicle’s desired route is not same as the platoon’s desired route; (2) the space gap

between the subject vehicle and the leading vehicle is too far; or (3) the subject vehicle’s desired

lane is not the current lane, due to the upstream congestion.

The CACC system determines the vehicle status based on the headway and the space gap

between two consecutive CAVs. The model requires that the two CAVs drive close enough to

maintain a platoon based on time headway or space headway. If the two consecutive CAVs are

not close enough to maintain a platoon, the following vehicle tries to join a platoon from a

significant distance. Figure 3 shows the CACC maneuvering algorithm.

29

CAV in CACC?

Headway<110% of desired headway & Space gap<110% of desired space gap

Speed < 120% of leader speedno

Acceleration = Leading vehicle acceleration +

1.64ft/s2

yesno

yes

no

Acceleration = use car-following model

with desired CAV headway

Headway=desired CAV headway & speed=leader speed

Acceleration = leader acceleration

Headway > desired CAV headway

Acceleration = Leader acceleration

* 1.25

yes

no

yes no

Leading vehicle accelerates?

Acceleration = Leader acceleration /

1.25

yesno

Maintain platoon

Join Platoon

Figure 3. CACC Maneuvering Algorithm. CACC = cooperative adaptive cruise control; CAV = connected

automated vehicle.

Vehicle Lateral Movement

Vehicle Platooning

Vehicle platooning is defined as a group of vehicles that travel closely together in an

actively coordinated formation. Previous research has shown that vehicle platooning increases

fuel and traffic efficiency, safety, and driver comfort (Bergenhem et al., 2012; Kanaris et al.,

1997). CACC and vehicle platooning have many things in common. Both strategies try to make

a platoon with adjacent vehicles to reduce headway and increase capacity. However, the

30

difference between CACC and platooning is the vehicle’s lateral movement; vehicle platooning

changes vehicles’ lanes, but CACC does not.

Platooning Algorithms

Any independent CAV (a CAV not in a platoon) can join a vehicle platoon. The CAV

tries to find a platoon in an adjacent lane and tries to join the platoon. If the CAV finds the

platoon on the left and the CAV itself is independent, the CAV tries to conduct a platooning

maneuver. If the CAV could not find the platoon in the left lane, the CAV searches for a platoon

on the right. The vehicle searches adjacent lanes until a potential platoon is found. If there is a

platoon on these far-left or far-right lanes, the CAV tries to change lanes to the left or right.

Once the located platoon is to the lane immediately left or right of the CAV, the CAV can then

initiate a platooning maneuver.

There are three different platooning maneuvers: joining the front of the platoon, joining

the end of the platoon, and joining the middle of the platoon. The platooning maneuvers are

determined by the location of the platoon and the CAV that wants to join the platoon. If a

platoon is behind the independent CAV on the right, the independent CAV can join to the front

of the platoon by changing lanes. If the independent CAV in the right lane is between the

platoon’s leader and the tail, the CAV may “consider” a cut-in maneuver. If the independent

CAV is behind the tail of the platoon, the CAV can join to the end of the platoon. Figure 4

shows examples of the platooning maneuvers.

Figure 4. Examples of Three Platooning Maneuvers. CAV_T = connected automated vehicle in the tail

position; CAV_M = connected automated vehicle as a platoon member; CAV_L = connected automated

vehicle as a platoon leader; CAV_I = connected automated vehicle entering vehicle.

31

The lane change occurs when safety can be guaranteed. The model determines safety

status by using the time headway between the independent CAV and the leading vehicle in the

platoon and the independent CAV and the following vehicle in the platoon (Desiraju et al.,

2015). The equations for lane changing are found in Desiraju et al. (2015). Based on the

Desiraju et al. study, this study adopted 1 second as a safe headway. In other words, if the time

headway between the subject vehicle and leading vehicle on the left (or right) is less than 1

second, the subject vehicle cannot change its lane. If the safe time headway is not guaranteed,

the model reduces the following vehicle’s acceleration to make a safe space. From the literature

review, the maximum number of vehicles in a platoon varies from 2 to 20 (Kesting et al., 2008;

Milanés and Shladover, 2014; Songchitruksa et al., 2016; Zhao and Sun, 2013). The maximum

platoon size of passenger vehicles for this study was set to 5 based on the previous referenced

research. Further, this study allowed forming a platoon with passenger CAVS and light truck

CAVS because the operational characteristics are similar. However, in the case of the HT

CAVs, the vehicle cannot form a platoon with passenger CAVs or light truck CAVs. In addition,

the number of vehicles in an HT CAV platoon is restricted to three for safety reasons. Figure 5

shows the platooning algorithm used in this study.

Simulation Scenario Development

This section describes the simulation testing plan used in the study, including the

development of the customized simulation tool used to integrate AVs and CAVs into the traffic

stream. Three different simulation scenarios were developed: a generic urban freeway segment

with a merge; a section of I-95 near Richmond, Virginia; and a segment of I-81 near Lexington,

Virginia. The developed algorithms were tested on the basic test network to ensure that the

driver behavior models were replicating the vehicle behavior as expected. Measures examined

included car-following behavior, merging behavior, speeds, and capacity values. For each of the

networks developed (Basic Freeway, I-81, and I-95), detailed validation procedures are provided.

Basic Freeway Segments

The basic freeway segments consisted of two networks that aimed to provide information

on the relationships among LV, AV, and CAV capacities. Figure 6a shows a test network and

the lane configurations for the simulation, where the ramp traffic must yield to the mainline

traffic. This network exists to show how much capacity would be able to flow when there is

little to no merging on the roadway. In this scenario, the ramp exists only to ensure (nearly) that

the occupancy of the roadway was near the highest possible. In this network, vehicles would

merge only if there was a significant gap in the traffic flow on the mainline. This closely

approximates full capacity without a bottleneck. Figure 6b shows the lane configurations for the

simulation, which simulates a forced merge scenario that could occur at the end of a ramp

acceleration lane (as modeled in this study) or a lane drop on a freeway segment. This model

will illuminate capacity as traditionally measured after a lane drop. The models were calibrated

by using values from the HCM for capacity in both scenarios for LVs and by output speed and

following distances for AVs and CAVs.

32

Veh Type is CAV and independent?

Left follower is in CACC and the CACC is not full

Left leader Timegap > Cut in Timegap_front

Move lane one leftActivate cut-in mode, left following vehicle

decelerates

yes

Do not change the lane

no

Left leader is in CACC and the CACC is not full

Left follower Timegap > Cut in Timegap_back

yesno

Right leader is in CACC and the CACC is not full

Right follower is in CACC and the CACC is not full

Two Left follower or leader is in CACC and the

CACC is not full

Two Right follower or leader is in CACC and the

CACC is not full

Right leader Timegap > Cut in Timegap_front

Right follower Timegap > Cut in Timegap_back

Move lane one rightActivate cut-in mode, right following vehicle

decelerates

yes

no yes

no

yes

yes

yes

yes

yes

yes

no

no yes

no

nono

no

Consider one lane left/right

Consider two lane left/right

Consider time headway and activate

cut-in mode

Figure 5. Platooning Algorithm. CAV = connected automated vehicle; CACC = cooperative adaptive cruise

control; Timegap_front = time gap in front of subject vehicle; Timegap_back = time gap behind the subject

vehicle in order to join platoon.

The input volume is 500 veh/hr to 6,000 veh/hr, incremented by 500 veh/hr every 5

minutes, and the on-ramp volume is 1,000 veh/hr. The merging segment in the basic freeway

segment scenario has no priority rules so that the mainline always maintains priority. After 1

mile, the throughput is collected to measure the capacity. The merging segment has one main

approach and one on-ramp approach with the addition of a 560-ft acceleration lane. The input

traffic is 500 veh/hr for both on-ramps and 500 veh/hr to 3,000 veh/hr, incremented by 500

veh/hr every 5 minutes, for the mainline. Unlike the basic freeway segment, VISSIM’s priority

33

rules for merging are implemented at the merge point to simulate forced merging behavior. On

both networks, traffic volume and density were collected every 500 ft. A total of eight

simulations per each scenario were conducted to account for randomness within the simulation.

Average values are reported in the “Results and Discussion” section. Default merging behavior

for ramps were used within VISSIM, using the ramp merging scenario option in the link

intersection settings.

Figure 6. Test Networks for Model Verification

I-81 Segment

The section of I-81 modeled in this study follows the Appalachian Mountains and has

varying grades. The study section is 7.8 miles long and is located between Exit 188B (US 60

East, Lexington) and Exit 180B (US 11, Natural Bridge). This section of roadway does not have

any interchanges between the start and finish of the segment. The section has a 31% truck

volume (VDOT, 2018b). Elevation data for I-81 were obtained from Google Earth and were

used to compute the grades. The segment has a downhill grade of -3.9% (0.8 miles) followed by

a 3.58% uphill (0.9 miles). Grades were relatively flat before and after the extended grade

section. The location of the I-81 scenario is shown in Figure 7.

The I-81 study area was modeled in VISSIM and used to simulate performance in the

presence of grades with consideration for HVs. INRIX data for the segment were used to set the

desired speed for the network calibration (70 mph). For the modeling of HVs, 12% LT and 88%

HT were entered (VDOT, 2018a). Power and weight parameters of HTs and LTs were modified

in VISSIM to account for the U.S. fleet characteristics.

The demand volume was set between 3,000 veh/hr and 6,000 veh/hr in each direction

with increments of 1,000 veh/hr (2,400-second total simulation with a 600-second warm-up

seeding period). Volume results were measured every 1,000 ft over nine simulation periods used

to estimate capacity.

34

Figure 7. I-81 Segment Location

I-95 Segment

A 15-mile segment of I-95 in Virginia between Richmond and Chesterfield was selected

to simulate LVs, AVs, and CAVs on an urban interstate segment. This corridor runs from Exit

74A at SR 195 (Downtown Expressway) to Exit 61 at Route 10, including seven interchanges.

The study focused on PM peak, from 4:45 PM to 5:45 PM. This network was calibrated from

another study conducted by a consultant. The calibration was conducted in accordance with the

guidelines in VDOT’s Traffic Operations and Safety Analysis Manual (TOSAM) (VDOT, 2015).

To be more specific, traffic volumes, speeds, and travel times were used for calibrating freeway

segments. Where the volume was more than 1,000 veh/hr, the calibrated threshold for traffic

volume was set to ±5%. In the case of speed, ±7 mph was used, and for the travel time, ±20%

was used for the calibration threshold. The research team confirmed that the volumes and speeds

were realistic for the simulation by comparing the volumes to those of VDOT count stations and

the speeds to INRIX data supplied by VDOT. The location of the I-95 scenario is shown in

Figure 8.

The study focused on the PM peak, from 4:45 PM to 5:45 PM. For the simulation, the

study had 30 minutes of the warm-up period and 2 hours of the peak period, including the peak

hour. Also, simulation results were collected by 15-minute periods. The demand for this

network was from the calibrated model with observed variation in the peak period timeframe.

Table 11 shows the simulation periods for this study, with analysis being conducted in the peak

hour only.

35

Figure 8. I-95 Segment Location

Table 11. Simulation Periods for I-95

Warm-up Period Peak Period Network Peak Hour

3:30 to 4:00 4:00 to 6:00 4:45 to 5:45

Simulation Plan

The simulation plan was based on baseline scenarios that were implemented only on the

basic freeway segments. In the other two scenarios, the plan was modified as stated here.

36

Vehicle Penetration Rates and Baseline Scenarios

The MP is a percentage of vehicles on the roadway that can be classified as AVs, CAVs,

or LVs. AV and CAV penetration rates were tested at varying levels. Similar efforts have been

investigated for AVs and CAVs (Shladover et al., 2012; Zhao and Sun, 2013).

Table 12 illustrates a total of 21 MP levels used in this study, in which emphasis was

placed on AVs and CAVs, but the anticipated low penetration in the early stages of deployment

was also considered. When AV and CAV MP levels were considered, an equal distribution of

the various determined vehicle behavior models were considered to investigate the mixed effects

of AVs or CAVs with different operational parameters from Table 8.

In addition, for the HV simulation, buses and single-unit trucks were treated as LTs and

trailer trucks were treated as HTs as specified in the latest HCM (Transportation Research Board,

2016). Based on volume and classification data from Virginia interstate highways (VDOT,

2018b), buses, two-axle trucks, and three-axle trucks were treated as a single-unit truck (LT), and

all trailers were treated as trailer trucks (HTs). The operational parameters of LTs can be found

in Table 9, and the operational parameters of HTs can be found in Table 10.

Table 12. Market Penetration Distribution

MP Scenario LV AV CAV

0 100% 0% 0%

1 80% 20% 0%

2 80% 0% 20%

3 60% 40% 0%

4 60% 20% 20%

5 60% 0% 40%

6 40% 60% 0%

7 40% 40% 20%

8 40% 20% 40%

9 40% 0% 60%

10 20% 80% 0%

11 20% 60% 20%

12 20% 40% 40%

13 20% 20% 60%

14 20% 0% 80%

15 0% 100% 0%

16 0% 80% 20%

17 0% 60% 40%

18 0% 40% 60%

19 0% 20% 80%

20 0% 0% 100%

MP = market penetration; LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles.

37

Baseline Scenarios

The baseline scenarios are provided here for each of the three simulations conducted in

this study. The details provided are the percentages of LVs, AVs, and CAVs; the amount of

HVs; and the amount of demand on the roadways.

Basic Freeway Segments

The baseline scenario for this simulation was 100% LV with VISSIM standard driver

behavior. The input demand for this was set to a very high value to ensure that there would be

much more demand than the roadway could accommodate in order to measure capacity. There

were scenarios that varied the amount of LVs, AVs, and CAVs that are detailed in Table 13.

These scenarios were expanded in order to take into consideration the type of vehicle movement

aggressiveness, which is detailed in Table 14. The scenarios shown in Table 13 were then

expanded to account for 5%, 10%, and 15% HV in those scenarios. The results of the HV

simulations are provided in Table 15. In addition, all the simulations used intermediate

aggressiveness parameters from Table 8.

Table 13. Capacity Simulation Results

Scenario No.

Market Penetration (%) Capacity (veh/hr/ln)

LV AV CAV Basic Freeway Segment Merging Freeway Segment

100 100 0 0 2286 1753

101 80 20 0 2410 1863

102 80 0 20 2477 1979

103 60 40 0 2524 1930

104 60 20 20 2581 1999

105 60 0 40 2746 2139

106 40 60 0 2653 2186

107 40 40 20 2723 2217

108 40 20 40 2906 2270

109 40 0 60 3139 2264

110 20 80 0 2789 2340

111 20 60 20 2855 2245

112 20 40 40 3041 2371

113 20 20 60 3322 2358

114 20 0 80 3757 2505

115 0 100 0 2942 2602

116 0 80 20 3029 2597

117 0 60 40 3241 2483

118 0 40 60 3523 2620

119 0 20 80 3963 2682

120 0 0 100 4373 2802

LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles.

Intermediate driver behavior was used.

38

Table 14. Results of the Capacity Simulations With Different Aggressiveness Settings (veh/hr)

Market Penetration

(%)

Intermediate

Aggressive

Conservative

Manufacturer

Liability

LV AV CAV Basic Merging Basic Merging Basic Merging Basic Merging

100 0 0 2286 1753 - - - - - -

80 20 0 2410 1863 2565 1924 2219 1824 2172 1799

80 0 20 2477 1979 2616 1934 2241 1900 2175 1882

60 40 0 2524 1930 2853 1930 2184 1983 2034 1871

60 20 20 2581 1999 2905 2063 2208 2013 2035 1874

60 0 40 2746 2139 3071 2141 2270 2005 2046 1892

40 60 0 2653 2186 3196 2089 2188 2091 1912 1873

40 40 20 2723 2217 3251 2158 2218 2079 1917 1846

40 20 40 2906 2270 3441 2323 2283 2041 1929 1821

40 0 60 3139 2264 3654 2397 2414 2052 1944 1827

20 80 0 2789 2340 3630 2262 2226 2214 1798 1820

20 60 20 2855 2245 3674 2313 2251 2117 1803 1798

20 40 40 3041 2371 3847 2473 2315 2130 1810 1753

20 20 60 3322 2358 4188 2517 2441 2165 1825 1776

20 0 80 3757 2505 4407 2556 2618 2232 1846 1800

0 100 0 2942 2602 4181 2267 2320 2309 1686 1754

0 80 20 3029 2597 4246 2305 2359 2232 1693 1721

0 60 40 3241 2483 4345 2726 2437 2154 1705 1708

0 40 60 3523 2620 4379 2878 2552 2244 1722 1687

0 20 80 3963 2682 4391 3056 2696 2306 1742 1722

0 0 100 4373 2802 4393 2933 2859 2408 1762 1762

LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles; - = not applicable.

The driver behavior noted in the columns were for AV and CAV only. LV used VISSIM driver behavior

distributions.

I-81 Scenarios

Scenarios varied by vehicle composition with HV percentages of 0%, 30%, and 50%. In

each of the three HV scenarios, MPs of AVs and CAVs were increased by intervals of 20%,

resulting in 21 MP scenarios per case (LV, AV, CAV combinations, as shown in Table 12).

Each HV percentage contained LTs and HTs in accordance with the traffic count data (VDOT,

2018a). In addition, the operational parameters for LTs and HTs were adopted from Tables 9

and 10, respectively. Scenarios were numbered 100 to 120 (for 0% HV), 300 to 320 (for 30%

HV), and 500 to 520 (for 50% HV).

I-95 Scenarios

The simulation was conducted with the current demand as a baseline. Demand was

increased to 150% and 200% for all ramps and mainline inputs to the simulation of the current

demand to understand capacity better and to see how increasing demand in future years would be

affected by the AV and CAV technologies.

39

Table 15. The Simulation-Based Capacities With Heavy Vehicles (veh/hr)

Market Penetration

(%)

Passenger

Vehicles

5% HV

10% HV

15% HV

LV AV CAV Basic Merging Basic Merging Basic Merging Basic Merging

100 0 0 2286 1753 2209 1893 2125 1829 2047 1771

80 20 0 2410 1863 2300 1954 2190 1897 2106 1829

80 0 20 2477 1979 2340 2067 2221 2028 2119 2025

60 40 0 2524 1930 2393 2026 2282 1978 2196 1966

60 20 20 2581 1999 2440 2143 2324 2142 2220 2088

60 0 40 2746 2139 2558 2319 2424 2208 2328 2205

40 60 0 2653 2186 2502 2078 2385 2089 2273 2025

40 40 20 2723 2217 2533 2224 2416 2173 2298 2132

40 20 40 2906 2270 2651 2312 2537 2270 2405 2252

40 0 60 3139 2264 2879 2378 2724 2365 2554 2381

20 80 0 2789 2340 2603 2300 2496 2191 2393 2111

20 60 20 2855 2245 2641 2241 2532 2252 2433 2217

20 40 40 3041 2371 2803 2366 2648 2377 2530 2347

20 20 60 3322 2358 3022 2437 2842 2581 2670 2495

20 0 80 3757 2505 3393 2561 3148 2714 2960 2622

0 100 0 2942 2602 2725 2270 2608 2216 2486 2212

0 80 20 3029 2597 2778 2535 2649 2339 2525 2298

0 60 40 3241 2483 2939 2559 2772 2489 2654 2438

0 40 60 3523 2620 3187 2593 2970 2659 2792 2579

0 20 80 3963 2682 3600 2638 3339 2788 3090 2798

0 0 100 4373 2802 4127 2886 3773 2947 3506 2936

HV = heavy vehicles; LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles.

The driver behavior noted in the columns were for AVs and CAVs only. LVs used VISSIM driver behavior

distributions.

The scenarios were considered with 100% AV and 100% CAV only. Truck volumes

were converted to passenger cars by the HCM method of passenger car equivalents. In addition,

all the simulations used intermediate aggressiveness parameters from Table 8.

Analysis of Test Networks

Performance Metrics

Consideration of other measures of effectiveness (MOEs) helps shape a picture of how

changes in traffic composition will affect operations on interstate highways. Capacity is widely

used to demonstrate the effect of AV and CAV technologies on traffic performance (Arnaout and

Bowling, 2011; Ntousakis et al., 2015; Shladover et al., 2012; VanderWerf et al., 2001).

Shladover et al. (2012) demonstrated how a 100% CACC penetration rate could increase the

capacity up to 4,000 veh/hr/ln.

Other MOEs also appeared in the literature. These MOEs can be classified into

macroscopic and microscopic measures. Under the macroscopic measures, speed and density

have been evaluated (Melson et al., 2018; Milanés and Shladover, 2014; van Arem et al., 2006).

40

Under the microscopic measures, factors such as travel time (Arnaout and Arnaout, 2014;

Kesting et al., 2008) and delay (Bierstedt et al., 2014) were used. This study concentrated on

throughput and speed as MOEs.

Sample Size Calculations

The sample size for the number of simulations needed for each case study was computed

based on the Federal Highway Administration (2016) equation (Eq. 14) that estimates the

minimum required number of replications of a simulation:

𝑁𝑚𝑖𝑛 = (𝑡𝑛−1,95%𝑠

𝑒��)2 (𝐸𝑞. 14)

where

𝑁𝑚𝑖𝑛 is required number of model runs

n is number of initial model runs (i.e., 4)

��, 𝑠 is mean and standard deviation of the initial runs

𝑡𝑛−1,95% is t statistic for n-1 degrees of freedom and 95% confidence level

𝑒 is tolerance error.

Tolerance error, e, is the maximum of the two calculated tolerance errors at the two

critical time intervals. It is computed in accordance with Equation 15 as follows:

𝑒 =

𝑡𝑛−1,95%𝑠

√𝑛��

(𝐸𝑞. 15)

For each scenario in each of the three case studies, the two critical periods were used to

compute the tolerance error. The critical period was assumed to correspond to the time intervals

resulting in the two most significant throughput values.

The required sample size was computed for each scenario, and the selected sample size

for the case study was the maximum of the obtained required sample sizes. For instance, for I-

81’s link 27, scenario 504 resulted in the largest required sample size, which turned out to be 7.

Then, for the entire sample consisting of four random seed results at the critical time

periods:

�� = 3271.59

𝑠 = 131.51

E = max (0.0169, 0.0074) = 0.0169 (smaller than 5%) use 0.05.

𝑁𝑚𝑖𝑛 = (𝑡𝑛−1,95%𝑠

𝑒��)2 = (

3.182∗131.51

0.05∗3271.59)2= 6.54; round up to 7.

41

After the first four runs for each of the three case studies were conducted, the sample size

was computed. The basic highway segment required four runs (but five were modeled), the

urban interstate required nine, and the rural interstate required seven.

Analysis Methods

This study aimed to compare the impacts on operations under different technology

distributions. For that reason, when the outcomes of the MOEs were compared, a paired two-

tailed t-test was performed to determine whether or not the difference between results was

significant. The assumption was that p-values smaller than 0.1 caused a rejection of the null

hypothesis (which assumes equal sample means) or that compared samples were significantly

different. The t-tests were chosen to contrast different scenarios with similar characteristics,

varying one variable.

Paired Sample t-Test

A paired sample t-test is a statistical method used to compare two sample means. The

goal is to determine whether or not the mean difference between the two sets of results is zero.

In a paired t-test, the null hypothesis will be that the true mean difference is zero. Thus, if the

null hypothesis is rejected, the means from the two samples are considered to be significantly

different. In order to accept or reject the null hypothesis, a p-value for the test statistic is

obtained. The test statistic is computed in accordance with Equation 16 as follows:

𝑡𝑜𝑏𝑠 =𝑦𝑑

𝑠𝑑

√𝑛⁄

(Eq. 16)

where

𝑦𝑑 is the sample mean difference (average of the differences between each pair of

observations from the two compared samples)

𝑠𝑑 is the sample standard deviation for the differences

N is the sample size.

The t-distribution tables were then used to compare the obtained 𝑡𝑜𝑏𝑠 value to the tn−1

distribution. This gave the p-value for the paired t-test, which is the probability of finding the

observed or more extreme results when the null hypothesis is confirmed. With a p-value greater

than 𝛼 (assumed to be 0.1 in this study), the null hypothesis is not rejected (equal sample means)

and there is insufficient evidence to suggest that results from the scenarios are significantly

different. The comparison for this research was the average of the eight simulations using the

maximum 1 hour volume as the unit of measure.

42

RESULTS AND DISCUSSION

Corridor Simulation Networks

Model Verification

The information in the following sections pertains to information for the non-merging

scenario. The merging scenario was verified using the knowledge of the research team and staff

of the Virginia Transportation Research Council (VTRC) to verify that the model was producing

results that were realistic for LVs, AVs, and CAVs. The theoretical values shown here are with

headways that do not reflect safety constraints seen in field implementation of vehicles. These

capacity values are theoretical upper bounds for capacity, which are much higher than field-

measured values.

LVs

As discussed previously, LVs represent human-driven vehicles without any automation

features. From the previous studies listed in Appendix A, the headway (Tn) of the LV was

chosen as 1.41 seconds. In an ideal case, the capacity of the single-lane road can be calculated

by Equation 17. In this case, the capacity of LVs will be 2,553 veh/hr/ln. This capacity is for the

basic freeway segment; a merging scenario would reduce the capacity based on the vehicle

interactions.

𝐶 = 3600

𝑇𝑛 (𝐸𝑞. 17)

AVs

As discussed previously, the headway of AVs was chosen to be 1.07 seconds (see Table

3) and was governed by the same car-following model as for LVs. AVs may have additional

automated driving features such as lane-keeping assistance, emergency braking, or parking

assistance, but those applications do not affect the theoretical maximum road capacity where

vehicles do not change lanes. The only factor that affects road capacity is the desired speed, and

all AVs have the same desired speed equivalent to the speed limit. By Equation 13, the capacity

of AVs in an ideal case is 3,365 veh/hr/ln. This capacity is for the basic freeway segment; a

merging scenario would reduce the capacity based on the vehicle interactions. It is expected that

the merging scenarios for AVs would be more orderly and would have a higher capacity.

CAVs

CAVs are AVs with communication capability. CAVs can generate CAV platoons and

reduce headways by communicating their intentions and actions with each other directly and

with low latency. Each CAV platoon consists of a leader vehicle, a tail vehicle, and (if there are

more than two vehicles) a member vehicle. The leader represents the first CAV in a platoon and

43

follows AVs or LVs and therefore must use AV headways. The member and tail vehicles follow

other CAVs and may use CAV headways, which are set at 0.65 seconds. Detailed characteristics

are discussed in the next section. The capacity of the CAV is determined by the maximum

platoon size, headway of the AV for the CAV leader, and headway of the CAV for CAV

members. Equation 18 shows the theoretical capacity of a CAV. According to the equation,

when the maximum platoon size is five, the capacity is 4,904 veh/hr/ln. This calculation

assumes no HVs and 100% of vehicles in platoons.

𝐶 = 3600𝑃𝑚𝑎𝑥

𝑇𝐴𝑉 + (𝑃𝑚𝑎𝑥 − 1)𝑇𝐶𝐴𝑉 (𝐸𝑞. 18)

where

𝑃𝑚𝑎𝑥 is the platoon size (number of vehicles).

Simulation Plan

Basic Highway Segment

The capacities of LVs, AVs, and CAVs were examined on the sample networks.

Twenty-one scenarios, numbered 100 to 120, were developed to test different MP rates for each

vehicle. Table 12 shows the scenarios tested. The operational parameters in Table 8 were used

for measuring the capacity for the passenger LVs, AVs, and CAVs. The results for each scenario

are expressed as an average of the results from the number of repetitions calculated from the

sample size calculation provided previously.

Scenario 100, which represents the LV only scenario, shows a theoretical maximum

capacity of 2,553 veh/hr/ln; however, the capacity from the basic freeway segment is 2,286

veh/hr/ln and the capacity from the merging freeway segment is 1,753 veh/hr/ln. The difference

between the capacities for the basic freeway segment and the merging freeway segment comes

from the gap due to the merging process. The HCM states that the capacity under the queue

discharge situation is substantially below the basic highway capacity of 2,400 veh/hr/ln

(Transportation Research Board, 2016). The simulation adopted different operational parameters

for each vehicle to mimic real traffic flow. Therefore, the observed capacity was less than the

theoretical capacity. Indeed, the capacity of the basic freeway segment from the latest HCM

showed 2,400 veh/hr/ln when the segment’s free-flow speed was 75 mph (Transportation

Research Board, 2016). Therefore, the capacity from the simulation seems reasonable. In Figure

9, the capacity is shown between a density of 40 to 50 vehicles per mile. Oversaturation was not

observed in the basic freeway segment since there was no delay downstream. However, in the

case of the merging segment, the lane before the merge point had congestion as needed to

measure the capacity at the merge point.

44

Figure 9. Density-Volume Plot for LV Only Scenario. LV = legacy vehicles.

In the theoretical case for AVs, the capacity was calculated as 3,365 veh/hr/ln.

According to scenario 115, which represents the AV only scenario, the capacities were 2,942

veh/hr/ln for the basic freeway segment and 2,602 veh/hr/ln for the merging freeway segment.

In both segments, there was merging behavior that would reduce capacity. The merging freeway

segment, being 700 less than the theoretical, showed a similar percentage reduction in the

theoretical capacity from the LV only scenario. Similarly, the theoretical capacity of CAVs was

4,904 veh/hr/ln; the measured capacity of the basic freeway segment was 4,373 veh/hr/ln; and

the capacity of the merging freeway segment was 2,802 veh/hr/ln. Based on the simulation

results and confirmed from previous studies, the model outputs were reasonable since most of

the studies revealed the capacity of 100% CACC vehicles to be near 4,000 veh/hr/ln. Figures 10

and 11 show that AVs and CAVs can sustain higher densities than LVs because all AVs and

CAVs have the same desired speed and can achieve the speed precisely due to the computer

controlling the driving behavior.

45

Figure 10. Density-Volume Plot for AV Only Scenario. AV = automated vehicles.

The simulation shows that the introduction of AVs could increase the capacity even at

low penetration rates. Compared with 100% LV, traffic consisting of 80% LV and 20% AV

would increase capacity by 5.4% at basic freeway segments and 6.3% at merging freeway

segments. The effects of CAVs at low penetration rates were larger than for AVs. When the

percentage of LVs was 80% and the percentage of CAVs was 20%, the capacity was increased

by 8.3% at the basic freeway segments and 12.9% at the merging segments. In the case of higher

penetration rates of AVs and CAVs, the increased capacity was significant. With 100% AV, the

capacity increased 28.7% at the basic freeway segments and 48.4% at the merging freeway

segments. With 100% CAV, the capacity increased 91.3% at the basic freeway segments and

59.8% at the merging segments. The merging segment capacity shows less of an improvement

with higher CAV penetration rates when compared to basic freeway segments because CAV

platooning does not occur in merging segments and cannot improve capacity.

46

Figure 11. Density-Volume Plot for CAV Only Scenario. CAV = connected automated vehicles.

Table 13 shows the effect of the vehicle fleet mix on capacity from the basic highway

segment simulations.

Aggressiveness Levels

The previous simulations tested LVs, AVs, and CAVs with the same driving

characteristics, such as desired speed, desired time headway, and acceleration/deceleration

capability. One likely scenario is that next-generation vehicles will have varying driving

characteristics based on manufacturer preferences and vehicle type. Also, it is expected that

vehicles will have passenger comfort settings for speed and aggressiveness, much like the ACC

systems do today. In order to test these different parameters, four different levels of operational

parameters—aggressive, conservative, intermediate, and manufacturer liability—were examined

47

on the test network. Table 14 shows the capacities for different levels of operational parameters

assuming all vehicles are using the aggressiveness settings shown in the column of the table.

The specific operational parameters for each vehicle will be decided by the passenger and

the manufacturer, with boundary conditions determined by law. The passengers may select the

operational parameters that mimic human driving and feel familiar. These parameters may be

similar to the intermediate or conservative settings. Manufacturers may be incentivized to avoid

crashes and the resulting civil liability and may select the manufacturer liability scenarios.

The aggressive scenario increased the capacity up to 42% at the basic freeway segment

and increased it up to 13.9% at the merging freeway segment. However, in some merging cases,

the aggressive driving decreased capacity because the aggressive driving interrupted smooth

merging. However, conservative driving decreased the capacity of the roadway up to 34.6% at

the basic freeway segment and up to 14.4% at the merging freeway segment based on the 100%

LV scenario. Likewise, under the manufacturer liability case, capacity reductions reached 59.7%

at the basic freeway segment and 37.1% at the merging freeway segment compared to the 100%

LV scenario.

The benefits from the aggressive scenarios were achieved for combinations with high AV

percentages. CAVs did not benefit as much from aggressive scenarios because the coordination

between vehicles produced efficient results under intermediate conditions due to their ability to

form platoons. CAVs were also the most negatively affected by more conservative parameter

settings with increased headways.

HVs

HV proportions of 5%, 10%, and 15% were considered for simulation because the

average percentage of HVs on Virginia interstate highways in 2018 was 11.6% (VDOT, 2018b).

The aggressiveness was set at the intermediate level for these simulations. In each scenario, the

HVs consisted of 18% LT and 82% HT, based on the 2018 Virginia annual average daily traffic.

Simulations were conducted on the same network with the same input volume. Each scenario

was simulated 5 times with different random seeds to produce an average capacity value. AVs

and CAVs were distributed at the same percentages for HVs and passenger vehicles.

Table 15 shows the capacities for different percentages of HVs. The capacity trend in the

presence of HVs is similar to that of passenger vehicles: a larger percentage of AVs or CAVs

achieves a higher capacity. The introduction of HVs, however, resulted in a decrease in capacity

across all combination scenarios when compared to passenger vehicles only. The reduction in

capacity increased with increasing HV percentage. In the case of the merging freeway segment,

the lower acceleration and deceleration rates of HVs produced larger gaps during the frequent

stop-and-go traffic and therefore lower capacities.

48

T-tests were conducted for all scenarios. All scenarios were calculated to be statistically

significant at the 90% confidence interval. The detailed scenario descriptions for the HV

simulations are provided in Appendix B, and the t-test results are provided in Appendix C.

Scenarios 200 to 220 represent 5% HV, scenarios 300 to 320 represent 10% HV, and scenarios

400 to 420 represent 15% HV.

Analysis of Test Networks

This section provides the analysis and discussion for the I-95 and I-81 test networks.

Different tests were conducted on each network to understand how AVs and CAVs responded to

the unique characteristics of each network. The I-95 tests focused on congestion due to traffic

demand, and the I-81 scenario focused on the impact of trucks. Detailed statistical results for all

scenarios on both networks are provided in Appendix C.

I-95 Network

An analysis of throughput and speed on the I-95 corridor was conducted with 100% LV,

100% AV, and 100% CAV. The intermediate car-following behavior was implemented for the

AV and CAV scenarios.

Speed

Speed is one of the MOEs used to evaluate the impact of AVs and CAVs on the

operations of the highway. Figures 12 through 14 show that at low demand, especially in the

100% demand scenarios and in the off-peak northbound direction, the average speed of LVs is

usually higher than that of AVs or CAVs because LVs can drive at the free-flow speed, which is

usually above the posted speed limit. In contrast, AVs and CAVs are programmed to follow all

traffic rules; therefore, they could not drive above the posted speed limit. At more congested

segments, AVs showed the most significant improvements in speed results over LVs. Speed

increased by up to 18% during the 150% demand case and up to 14% during the 200% demand

case.

49

Figure 12. 100% Demand Average Speed. SB = southbound; LV = legacy vehicles; AV = automated vehicles;

CAV = connected automated vehicles; NB = northbound.



50



The speed heatmaps provided in Figures 15 through 17 show the average speed at each

segment at each 15-minute evaluation period during the peak period. In these heatmaps, each

small square represents a 100-ft segment of roadway, and the period is 15 minutes. In the

uncongested segments, the LV speed was higher than the AV or CAV speed because the

simulation allowed for LVs to drive above the speed limit. In the congested periods, LVs drove

slower than AVs or CAVs; however, in uncongested areas, the simulation allowed for “human”

drivers to exceed the speed limit where the AVs were assumed to obey the speed limit strictly.

CAVs showed lower speeds than AVs, because of the platooning process. The platooning

process of CAVs was beneficial in the case of basic freeway segments, but for complex

segments, such as on- and off-ramp and weaving segments, AVs and CAVs may not be as

efficient as human drivers due to safety rules and less ability to adapt. These segments reduce

platoon speeds as vehicles maneuver into and out of the platoon, and this affects the speed of all

of the following vehicles due to their short headways. In addition, at the weaving segment, every

independent CAV will attempt to join or form a platoon, and this additional lateral movement

can impede the movement of non-platooned vehicles. In the case of AVs, their desired speeds

are the same as CAVs, but they do not need to change their lane for platooning so that vehicles

can drive with a small amount of interaction with other vehicles. The northbound heatmaps also

show similar results.

51

Figure 15. 100% Demand, Speed Heatmap for I-95. LV = legacy vehicles; AV = automated vehicles; CAV =

connected automated vehicles.

Figure 16. 150% Demand, Speed Heatmap for I-95, 150% of Traffic Demand. LV = legacy vehicles; AV =

automated vehicles; CAV = connected automated vehicles.

52

Figure 17. 200% Demand, Speed Heatmap for I-95, 200% of Traffic Demand. LV = legacy vehicles; AV =


Throughput

Throughput, measured in vehicles per hour per lane, was used to evaluate the impacts of

AVs and CAVs on I-95. The initial analysis in this network was based on the input volumes

from the calibrated I-95 model. After the base simulations were conducted, the input volume

was increased to 150% and then 200% of the current traffic demand.

Throughput results were collected for both the southbound and northbound directions.

For the southbound direction, which was the peak direction during the PM peak period,

throughput results for the 100% demand scenarios (current day demand) for LVs, AV%, and

CAVs are presented in Figure 18. On all segments, the 100% LV scenario showed the best

performance compared to the 100% AV and 100% CAV scenarios. In the case of the

southbound direction, AVs and CAVs showed 4% and 6% lower throughput than LVs; in the

case of the northbound direction, AVs and CAVs showed 1% and 5% lower throughput than

LVs, respectively. Notably, the 100% CAV scenario showed lower throughputs, which was the

opposite of the test network result. The reason is that when the platoons of vehicles enter

sections of roadway where there are weaving sections, the flow of vehicles becomes unstable as

vehicles seek to join platoons. This could be a limitation of the modeling strategy as CAVs

would most likely not try to form platoons in weaving areas. However, the result is interesting

as it shows a weakness of the CAV technology. The throughput per lane is far below the

highway capacity for these vehicles because of the weaving section.

53

Figure 18. 100% Demand Southbound (SB) and Northbound (NB) Throughput. LV = legacy vehicles; AV =


Figure 19 shows the results of the 150% traffic demand. Congestion starts to form, and

the benefits of AVs and CAVs over LVs are seen. The peak southbound direction showed more

significant benefits of AVs and CAVs as compared to the northbound direction, which had less

congestion. The AV and CAV scenarios both increased throughput in the southbound direction

as compared to the 100% LV scenario. However, the AV scenario resulted in the best

performance, in this case, reaching a 10% increase in throughput along the Belles Rd. to SR-150

segment.

Figure 20 shows that at 200% demand, CAVs proved to be more beneficial. Under more

congested conditions, CAVs, as compared to LVs, produced a throughput increase of between

16% and 27% (298 to 818 veh/hr/ln increase). AVs, on the other hand, increased throughput

between 15% and 21% (284 to 594 veh/hr/ln increase). Similarly, in the northbound direction,

both AVs and CAVs proved to be more beneficial than LVs. AVs yielded better throughput in

the less congested segments, whereas CAVs gave the best performance with more congestion.

The percentage increase reached 21% for AVs and 25% for CAVs. These scenarios showed that

AVs are better in less congested conditions and CAVs are better in more highly congested

conditions.

54



The presence of weaving maneuvers at interchanges on I-95 showed the need for

specialized algorithms and processes for maneuvering vehicles in these complex areas.

Implementation of specialized algorithms to move platoons away from the weaving segments or

to make platoons break approaching a weaving segment could alleviate some of the throughput

issues seen in the simulation. OEMs are leveraging machine learning and artificial intelligence

research techniques to combat this challenging issue currently, and the research that will come

and policies from departments of transportation will inform how vehicles maneuver in these

complex areas.

Throughput results with increased demand of 150% and 200% show the benefits of AVs

and CAVs over LVs with increased congestion. CAVs proved to be most beneficial under the

more congested conditions with their increased ability to communicate and decrease unnecessary

stop-and-go occurrences, and AVs increased throughput in lightly congested conditions by

contributing to a smoother traffic flow. The results showed that the impact of the different

vehicle technologies is seen more clearly at the 200% demand level. At the 100% and 150%

demand levels, there were points along the corridor where there was not a significant difference

between the technologies on throughput.

55



I-81 Analysis

Capacity

For the I-81 segment, the simulation was conducted to understand the effect of grade, HV

technology, and AV technology on capacity. In this case study, the input volume was increased

from 6,000 to 12,000 with 1,000-veh/hr increments to obtain capacity results. Output volume

values were collected in VISSIM over nine 300-second time intervals every 1,000 ft, and the

maximum values were recorded to represent capacity.

Four data collection points were considered for the analysis. The first data collection

point (DC1) was located on the flat section preceding the extended grade section. The second

point (DC2) was located on the downhill segment, DC3 was located on the uphill segment, and

DC4 measured the volume on the flat section following the uphill grade. Figure 21 shows the

locations.

Figure 21. Elevation Profile and Data Collection Points

56

Results showed a decrease in capacity for all vehicles on the downhill segments because

vehicles made maneuvers due to the change in speed of the trucks. This behavior was confirmed

by the research team’s observations of the roadways. The downhill grade affected the lighter

cars differently than the HTs where the mass of the vehicle often caused the HTs to move faster

than the passenger cars. The simulation results that showed turbulent traffic flow (and reduced

capacity) because of this phenomenon were confirmed by the research team’s anecdotal

experience driving through this segment of roadway. The uphill segment further reduced

capacity as HVs slowed on the uphill.

The scenarios in Appendices B and C labeled in the 100s correspond to the 0% HV cases,

in the 300s to the 30% HV cases, and in the 500s to the 50% HV cases. Similar trends of

capacity variation were observed for the segment for the different combinations. The values

themselves varied based on the modeled scenario. Table 16 shows capacity variations of the

100% LV, AV, and CAV scenarios along the data collection points. In Table 16, DC3 is

considered for the comparison of capacity values across scenarios.

A significant decrease in capacity was observed upon the introduction of HVs, as shown

in Table 17. The capacity reduction was observed at 36% from the 0% to the 30% HV scenario,

which accounts for the space the HVs occupy and the operational characteristics of the vehicles

creating uneven traffic flow. Between the 30% and 50% HV scenarios, capacity reductions were

still significant but did not exceed 14%. The highest capacity differences resulting from the

penetration of HVs occurred at lower LV and higher AV and CAV scenarios, especially CAVs

with reduced capabilities to form longer platoons.

AVs and CAVs were capable of improving capacity at any MP for all HV cases, as

shown in Table 18. Better results were obtained with LV percentages less than those for AVs

and CAVs. The best performance was achieved for the 100% CAV case, reaching 86%, 65%,

and 63% capacity increases, as compared to the 100% LV case, in the 0%, 30%, and 50% HV

scenarios, respectively.

Table 16. Capacity Differences Along Data Collection Points

% HV

Scenario

Capacity Difference (veh/hr/ln)

DC1-DC2 DC2-DC3 DC3-DC4

0% HV 100% LV 34.75 9.12 -17.11

100% AV 45.06 11.93 -29.89

100% CAV 65.27 14.71 -24.92*

30% HV 100% LV 33.14 18.01 -27.49

100% AV 18.61 18.6* -29.44

100% CAV 62.46 30.46 -26.94

50% HV 100% LV 15.92 26.03 -44.66

100% AV 25.63 16.06 -55.34

100% CAV 35.9 11.54* -91

HV = heavy vehicles; DC = data collection point; LV = legacy vehicles; AV = automated vehicles; CAV =

connected automated vehicles.

* P-value > 0.1, indicating no significant difference.

57

Table 17. Percentage Capacity Reductions With Increase in HV% (veh/hr/ln)

HV = heavy vehicles.

Table 18. Percentage Capacity Increases Due to AVs and CAVs

LV

AV

CAV

0% HV 30% HV 50% HV

Capacity

(veh/hr/ln)

%

Difference

From LV

Only

Capacity

(veh/hr/ln)

%

Difference

From LV

Only

Capacity

(veh/hr/ln)

%

Difference

From LV

Only

100% 0% 0% 2253 0% 1680 0% 1478 0%

0% 100% 0% 2912 29% 2003 19% 1752 18%

0% 80% 20% 2990 33% 2043 22% 1773 20%

0% 60% 40% 3190 42% 2152 28% 1843 25%

0% 40% 60% 3473 54% 2292 36% 1979 34%

0% 20% 80% 3880 72% 2501 49% 2166 46%

0% 0% 100% 4199 86% 2769 65% 2405 63%

HV = heavy vehicles; LV = legacy vehicles; AV = automated vehicles; CAVs = connected automated vehicles.

Speed

Speeds were also recorded and compared on the uphill data collection point. Results

showed increased speeds with increasing AV and CAV MPs. However, at higher LV

percentages (more than 60%), changes in speed were not significant, especially in the 0% HV

scenarios. Higher MPs of AVs and CAVs always resulted in better speed results. At lower

speeds, with increased percentages of HVs, AVs and CAVs were more capable of showing

significant speed increases (up to 49% for the CAV only case). Still, the highest speeds were

recorded for the highest CAV scenarios under any HV condition. The average speed values for

each scenario are shown in Table 19. CAV platooning raised the average speed significantly for

the entire traffic stream.

Compared Scenarios

% Difference Between

0% and 30%

Compared Scenarios


30% and 50% HV

100 vs. 300 25% 300 vs. 500 12%

115 vs. 315 31% 315 vs. 515 13%

116 vs. 316 32% 316 vs. 516 13%

117 vs. 317 33% 317 vs. 517 14%

118 vs. 318 34% 318 vs. 518 14%

119 vs. 319 36% 319 vs. 519 13%

120 vs. 320 34% 320 vs. 520 13%

58

Table 19. Speed Results Across Scenarios

LV

AV

CAV

0% HV 30% HV 50% HV

Speed

(mph)

% Difference

From LV Only

Speed

(mph)

% Difference

From LV Only

Speed

(mph)

% Difference

From LV Only

100% 0% 0% 63.99 0% 37.10 0% 35.89 0%

80% 20% 0% 64.10 0% 39.69 7% 36.65 2%

80% 0% 20% 64.08 0% 41.80 13% 38.39 7%

60% 40% 0% 64.23 0% 39.01 5% 39.21 9%

60% 20% 20% 64.37 1% 44.13 19% 40.22 12%

60% 0% 40% 64.37 1% 42.52 15% 42.39 18%

40% 60% 0% 64.76 1% 38.56 4% 39.10 9%

40% 40% 20% 64.46 1% 41.87 13% 42.40 18%

40% 20% 40% 64.72 1% 45.57 23% 42.49 18%

40% 0% 60% 64.92 1% 46.93 27% 44.56 24%

20% 80% 0% 65.42 2% 39.73 7% 40.43 13%

20% 60% 20% 65.65 3% 44.04 19% 41.85 17%

20% 40% 40% 65.54 2% 45.51 23% 43.20 20%

20% 20% 60% 65.65 3% 48.12 30% 45.63 27%

20% 0% 80% 65.06 2% 52.66 42% 48.47 35%

0% 100% 0% 69.96 9% 41.76 13% 41.16 15%

0% 80% 20% 69.70 9% 44.47 20% 43.06 20%

0% 60% 40% 69.66 9% 47.33 28% 46.39 29%

0% 40% 60% 69.20 8% 48.62 31% 46.52 30%

0% 20% 80% 69.29 8% 51.85 40% 48.68 36%

0% 0% 100% 69.92 9% 55.18 49% 52.70 47%

HV = heavy vehicles; LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles.

CONCLUSIONS

Virginia statutes do not explicitly prohibit platooning at close following distances. Virginia

does not specify any following distance or cut-in requirements for vehicles, relying instead

on the reasonable and prudent standard; therefore, Virginia comprises a permissive

regulatory environment for most platooning applications. The desired following distance

selected by AVs may vary based on a vehicle’s braking ability, assumptions about the

braking behaviors of other vehicles, interpretations of legal concepts such as the ACDA

doctrine, industry standards, consumer preference, recommended driver training distances, or

a range of other factors.

CAVs and AVs outperformed LVs in most scenarios, and CAVs outperformed AVs during

congestion. The operations of LVs, AVs, and CAVs at 100% MP were compared on a

selected segment of I-95 during the PM peak. The overall results showed a better

performance of AVs and CAVs as compared to LVs except in cases where LVs broke the

speed limit due to VISSIM driver behavior modeling. However, in congested conditions,

AVs presented a lower performance, with reduced speeds and increased travel times. AVs

have smaller headways than LVs, and under congestion, these short headways would

59

increase the frequency of stop-and-go occurrences and thus negatively affect speeds and

travel times. This issue of unnecessary stop-and-go due to short headways was fixed in the

case of CAVs because of their ability to communicate and increase string stability.

CAVs underperformed AVs in weaving scenarios on I-95. The results of the simulation

showed that CAV traffic flow is less stable than AV traffic flow in weaving areas. This

resulted in reduced speed and throughput in the weaving areas as compared to the basic

freeway segments. This shows that CAV algorithms will need to be flexible and smart in

order to maintain flow in urban areas with high amounts of weaving sections. Having

regulations that allowed for platoons moving to the left lane would allow for mitigation of

this phenomenon.

AVs and CAVs improve performance in corridors with extended grades and high truck

volumes. In the case of extended grades, and even in the presence of high percentages of

HVs, AVs and CAVs seemed to have a high potential of improving operations. AVs and

CAVs, even at low MPs, were still capable of mitigating the impact of grades on

performance. However, the best results were achieved at the highest MPs of AVs and CAVs

in the absence of LVs. Mostly, AVs and CAVs had similar impacts on improving the values

of the different MOEs such as speed and travel time, even in the presence of grades and HVs.

CAVs, however, particularly at high MPs, with their communication capabilities, were

capable of achieving the most capacity increases on the selected I-81 segment.

There are opportunities for significant capacity increases due to the introduction of AV

technology. These advantages are seen most in the CAV cases, where communication is

leveraged in order to gain safe and efficient car-following, which allows for the additional

capacity. The transition from LVs to CAVs could have stages where capacity is reduced,

however. These scenarios include (1) AVs where car-following is very conservative due to

liability concerns of the vehicle manufacturers and tier one suppliers, and (2) AVs that are

programmed for user optimal behavior, which could cause increased stop-and-go driver

behavior based on the programming of the automated driving algorithms.

RECOMMENDATIONS

1. VDOT’s Connected and Automated Vehicle Program Manager and VTRC should continue to

monitor AV car-following behavior models. As new developments occur, the information

should be shared with VDOT’s Traffic Engineering Division (TED) to ensure that VDOT

simulation models reflect the existing and anticipated vehicle fleet.

2. VDOT’s TED should use the capacities provided in this report as guidance when calibrating

models of CVs and AVs in simulations of freeway corridors. In cases where models are being

developed for future conditions, the information in this report can be used to provide

guidance on car-following behavior and expected capacity for calibration. Modifications to

60

VDOT’s Traffic Operations and Safety Analysis Manual (TOSAM) are not recommended at

this time but may be warranted in the future as more AVs become commercially available.

3. VDOT’s Connected and Automated Vehicle Program Manager and VTRC should investigate

methods to estimate proportions of vehicles on Virginia roads operating with connected and

automated functionality. The results of this study suggest that the impact of AVs and CAVs

on capacity is correlated with their MP, yet neither VDOT nor the Virginia DMV has a

procedure to estimate the proportion of vehicles with connected or automated capabilities.

Tracking sales or registration of vehicle models is inadequate, as after-market add-on or

over-the-air vehicle software updates may allow for increasing automation. This effort

should also consider surrogate measures such as user surveys, inspection data from other

states, federal data, and insurance industry data to develop estimates for Virginia.

4. VDOT’s TED and VTRC should investigate regulation of the operation of AVs and CAVs in

weaving areas. Additional research should be conducted to understand the operations of AV

movement in weaving areas. This research should survey OEMs for their implementation of

machine learning and artificial intelligence to assist in complex driving maneuvers.

IMPLEMENTATION AND BENEFITS

Implementation

For Recommendation 1, VDOT’s Connected and Automated Vehicle Program Manager

and VTRC are actively monitoring research developments in automated driving through

participation in the Transportation Research Board, the Connected Vehicle Pooled Fund Study,

NCHRP efforts, and a continuous review of the scientific literature. This effort will continue

with an emphasis on new developments in car-following and capacity modeling of CVs and AVs

based on empirical data.

For Recommendation 2, VDOT’s TED, when developing transportation models that

incorporate automated driving, will use the capacities described in this report when the modeled

conditions are similar to those in the study scenarios. These capacities can be used until they are

considered outdated and replaced by new capacities discovered as part of implementing

Recommendation 1 or a future study. In the next 2 years, VTRC will report on the potential to

incorporate the models developed in this study or in other future studies of VDOT’s Traffic

Operations and Safety Analysis Manual (TOSAM).

For Recommendation 3, VDOT’s Connected and Automated Vehicles Program Manager

and VTRC will investigate potential data sources, best practices, and potential policy that will

allow reasonable estimates of the percentage of vehicles on Virginia roads employing connected

and automated technologies. This might involve leveraging data sources from outside Virginia

such as insurance industry records, estimates from other states, manufacturer data, and survey

61

results. Alternatively, estimates could be obtained through policy changes and partnering with

other state agencies such as the Virginia State Police and the Virginia DMV. The selected

methodology can be used to develop an evolving estimate of both the percentage of vehicles on

the road with automated and connected functions and the precise capabilities of these

technologies and their rate of use. For example, although many cars may be equipped with

ACC, a smaller percentage are also equipped with lane-keeping technology, and an even lower

percentage of drivers may actually use these capabilities at any given time. The resulting

estimates can be used to validate capacity models developed as part of this study. VDOT may

begin to discuss policy changes in anticipation of greater capacities with increased CAV

penetration. The work to define this methodology should begin within 1 year of the publication

of this report.

For Recommendation 4, VTRC staff should work with the research team, other university

partners, industry partners, and VDOT staff to build a research statement to be voted on at the

next meeting of VTRC’s System Operations Research Advisory Committee. The research

statement should address AV operations in complex situations and have an outcome of informing

policy within Virginia, at the national level, and in AASHTO committees.

Benefits

The benefit of implementing Recommendation 1 is that leveraging new research on

forthcoming vehicle technologies ensures that VDOT uses the most accurate models of

autonomous vehicle behavior and capacity. With more accurate models, VDOT can make more

informed decisions on planning-level needs, investments in infrastructure, and freeway capacities

as the vehicle fleet continues to evolve with more automation and connectivity of the driving

task.

The benefits of implementing Recommendation 2 are more accurate models of freeway

capacity under an increasing AV and CV fleet. By implementing sophisticated capacity models

into long-range planning efforts, VDOT can make more informed decisions regarding future

infrastructure needs. By continuously revising, updating, and using these models, VDOT can

ensure that investment decisions rely on the latest validated estimates from empirical data. This

will produce second order benefits of reducing costs and risks while maximizing the benefits and

opportunities of vehicle automation.

The benefits of implementing Recommendation 3 are that VDOT will make better

decisions on investments in the rehabilitation, reconstruction, and building of new roadway

infrastructure. Capacity models developed in this study depend on the percentage of vehicles

with automation and connectivity and the characteristics of automated driving systems. With

estimates of CAV attributes and percentages in the fleet, VDOT will have more reliable capacity

estimates, which will allow VDOT to make more informed decisions on infrastructure

improvements needed to manage demand. This will allow VDOT to plan infrastructure

62

improvements to coincide with capacity changes due to increasing automation, thereby reducing

costs while managing congestion.

The benefits of implementing Recommendation 4 are that VDOT will be able to propose

regulation of vehicle operations in complex operational scenarios and will have a deeper

understanding of the machine learning and artificial intelligence used to inform driving decision

algorithms. This research could provide a greater opportunity for VDOT to work with OEMs to

ensure safe and efficient operation of AVs in Virginia. Any new regulations that would come

from this research should be developed in concert with OEMs and other transportation

stakeholders in Virginia.

ACKNOWLEDGMENTS

The researchers thank VTRC and VDOT for their continued support of this study.

Specifically, the authors thank the members of the technical review panel: Ken Earnest, Amanda

Hamm, Sanhita Lahiri, Hari Sripathi, Peng Xiao, and Mo Zhao. A special thanks goes to

Michael Fontaine for his support in the review of the final document and his valuable comments.

REFERENCES

Arnaout, G., and Bowling, S. Towards Reducing Traffic Congestion Using Cooperative

Adaptive Cruise Control on a Freeway With a Ramp. Journal of Industrial Engineering

and Management, Vol. 4, No. 4, 2011, pp. 699-717.

Arnaout, G.M., and Arnaout, J.-P. Exploring the Effects of Cooperative Adaptive Cruise Control

on Highway Traffic Flow Using Microscopic Traffic Simulation. Transportation

Planning and Technology, Vol. 37, No. 2, 2014, p. 186.

Audi AG. Owner’s Manual, 2017 Q7. Ingolstadt, Germany, 2016.

Bergenhem, C., Shladover, S., Coelingh, E., Englund, C., and Tsugawa, S. Overview of

Platooning Systems. In Proceedings of the 19th ITS World Congress, October 22-26,

Vienna, Austria, 2012.

Bierstedt, J., Gooze, A., Gray, C., Peterman, J., Raykin, L. and Walters, J. Effects of Next-

Generation Vehicles on Travel Demand and Highway Capacity. FP Think Working

Group, 2014, pp. 10-11.

Booz Allen Hamilton, WSP, and New Jersey Institute of Technology. NCHRP Report 891:

Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles.

Transportation Research Board, Washington, DC, 2018.

63

Buchwalter, J.D., Danne, W.H., Dvorske, J.J., Dietz, L.H., Eclaves, R.P., Esping, E.K., and

Larsen, S. American Jurisprudence 2d: Automobiles and Highway Traffic, Second

Edition. Jurisprudence Publishers, Inc., 1963.

Carbaugh, J., Godbole, D. and Sengupta, R. Safety and Capacity Analysis of Automated and

Manual Highway Systems. Transportation Research Part C: Emerging

Technologies, Vol. 6, Nos. 1-2, 1998, pp. 69-99.

Coppola v. Jameson (1972). 200 NW2d 877. Iowa Supreme Court.

Decker v. Wofford (1960). 104 NW2d 760. Michigan Supreme Court.

Delis, A., Nikolos, I., and Papageorgiou, M. Macroscopic Traffic Flow Modeling with Adaptive

Cruise Control: Development and Numerical Solution. Computers & Mathematics with

Applications, Vol. 70, No. 8, 2015, pp. 1921-1947.

Desiraju, D., Chantem, T., and Heaslip, K. Minimizing the Disruption of Traffic Flow of

Automated Vehicles During Lane Changes. IEEE Transactions on Intelligent

Transportation Systems, Vol. 16, No. 3, 2015, pp. 1249-1258.

Federal Highway Administration. NGSIM Fact Sheet: US Highway 101 Dataset. FHWA-HRT-

07-030. Washington, DC, 2007.

Federal Highway Administration. Simulation for Research on Automated Longitudinal Vehicle

Control. FHWA-PROJ-13-0030. Saxton Transportation Operations Laboratory, 2015.

Federal Highway Administration. Highway Performance Monitoring System: Field Manual.

Washington, DC, 2016.

Fernandes, P., and Nunes, U. Multiplatooning Leaders Positioning and Cooperative Behavior

Algorithms of Communicant Automated Vehicles for High Traffic Capacity. IEEE

Transactions on Intelligent Transportation Systems, Vol. 16, No. 3, 2015, pp. 1172-1187.

Fishelson, J., Freckleton, D., and Heaslip, K. Evaluation of Automated Electric Transportation

Deployment Strategies: Integrated Against Isolated. IET Intelligent Transport

Systems, Vol. 7, No. 3, 2013, pp. 337-344.

Huang, S., Ren, W., and Chan, S. Design and Performance Evaluation of Mixed Manual and

Automated Control Traffic. IEEE Transactions on Systems, Man, and Cybernetics—Part

A: Systems and Humans, Vol. 30, No. 6, 2000, pp. 661-673.

64

International Organization for Standardization. ISO 15622:2010(En): Intelligent Transport

Systems—Adaptive Cruise Control Systems—Performance Requirements and Test

Procedures. Geneva, Switzerland, 2010.

Kanaris, A., Ioannou, P., and Ho, F.-S. Spacing and Capacity Evaluations for Different AHS

Concepts. In Automated Highway Systems, Springer, Boston, 1997, pp. 125-171.

Kanellakopoulos, I., and Tomizuka, M. Commercial Trucks and Buses in Automated Highway

Systems. In Automated Highway Systems, Springer, Boston, 1997, pp. 213-246.

Kesting, A., Treiber, M., Schönhof, M., and Helbing, D. Adaptive Cruise Control Design for

Active Congestion Avoidance. Transportation Research Part C, Vol. 16, No. 6, 2008,

pp. 668-683.

Kesting, A., Treiber, M., and Helbing, D. Enhanced Intelligent Driver Model to Access the

Impact of Driving Strategies on Traffic Capacity. Philosophical Transactions of the

Royal Society A: Mathematical, Physical and Engineering Sciences, Vol. 368, No. 1928,

2010, pp. 4585-4605.

Le Vine, S., Kong, Y., Liu, X., and Polak, J. Vehicle Automation and Freeway ‘Pipeline’

Capacity in the Context of Legal Standards of Care. Transportation, 2017, pp. 1-30.

Le Vine, S., Liu, X., Zheng, F., and Polak, J. Automated Cars: Queue Discharge at Signalized

Intersections With ‘Assured-Clear-Distance-Ahead’ Driving Strategies. Transportation

Research Part C: Emerging Technologies, Vol. 62, 2016, pp. 35-54.

Leibowitz, H.W., Owens, D.A., and Tyrrell, R.A. The Assured Clear Distance Ahead Rule:

Implications for Nighttime Traffic Safety and the Law. Accident Analysis & Prevention,

Vol. 30, No. 1, 1998, pp. 93-99.

Li, J.-Q., Zhou, K., Shladover, S.E., and Skabardonis, A. Estimating Queue Length Under

Connected Vehicle Technology: Using Probe Vehicle, Loop Detector, Fused Data.

Transportation Research Record: Journal of the Transportation Research Board, No.

2366, 2013, pp. 17-22.

Melson, C.L., Levin, M.W., Hammit, B.E., and Boyles, S.D. Dynamic Traffic Assignment of

Cooperative Adaptive Cruise Control. Transportation Research Part C: Emerging

Technologies, Vol. 90, 2018, pp. 114-133.

Meissner, E., Chantem, T. and Heaslip, K. Optimizing Departures of Automated Vehicles from

Highways While Maintaining Mainline Capacity. IEEE Transactions on Intelligent


65

Milanés, V., and Shladover, S.E. Modeling Cooperative and Autonomous Adaptive Cruise

Control Dynamic Responses Using Experimental Data. Transportation Research Part C:

Emerging Technologies, Vol. 48, 2014, pp. 285-300.

National Academies of Sciences, Engineering, and Medicine. Implications of Connected and

Automated Driving Systems, Vol. 4: Autonomous Vehicle Action Plan. The National

Academies Press, Washington, DC, 2018.

Ntousakis, I.A., Nikolos, I.K., and Papageorgiou, M. On Microscopic Modelling of Adaptive

Cruise Control Systems. Transportation Research Procedia, Vol. 6, 2015, pp. 111-127.

Pearson, J.O. Automobiles: Sudden Emergency as Exception to Rule Requiring Motorist to

Maintain Ability to Stop Within Assured Clear Distance Ahead. American Law Reports,

75(Third), 2005, p. 333.

Raza, H., and Ioannou, P. Vehicle Following Control Design for Automated Highway

Systems. IEEE Control Systems Magazine, Vol. 16, No. 6, 1996, pp. 43-60.

Reiskin, J.S. FHWA Demonstrates 3-Truck Platoon in VA to Show Technology for Road, Fuel

Efficiency. Transport Topics, 2017.

Scribner, M. Authorizing Automated Vehicle Platooning: A Guide for State Legislators, 2018.

Shelton, J. Revolutionizing Our Roadways: Modeling the Traffic Impacts from Automated and

Connected Vehicles in a Complex, Congested Urban Setting. Texas A&M

Transportation Institute, College Station, 2016.

Shladover, S.E., Su, D., and Lu, X.-Y. Impacts of Cooperative Adaptive Cruise Control on

Freeway Traffic Flow. Transportation Research Record: Journal of the Transportation

Research Board, No. 2324, 2012, pp. 63-70.

Smith, B.W. Automated Vehicles Are Probably Legal in the United States. Texas A&M Law

Review 1, 2013, p. 411.

Songchitruksa, P., Bibeka, A., Lin, L.I., and Zhang, Y. Incorporating Driver Behaviors Into

Connected and Automated Vehicle Simulation. Center for Advancing Transportation

Leadership and Safety, 2016.

Spyropoulou, I. Modelling a Signal Controlled Traffic Stream Using Cellular Automata Models.

Transportation Research Part C: Emerging Technologies, Vol. 15, 2007, pp. 175-190.

66

Su, P., Ma, J., Lochrane, T., Dailey, D., and Hale, D. Integrated Adaptive Cruise Control Car-

following Model Based on Trajectory Data. In Proceedings of the Transportation

Research Board Annual Meeting, Washington, DC, 2016.

Talebpour, A., and Mahmassani, H. Influence of Connected and Autonomous Vehicles on

Traffic Flow Stability and Throughput. Transportation Research Part C: Emerging

Technologies, Vol. 71, 2016, pp. 143-163.

Teodorović, D., and Janić, M. Chapter 5—Capacity and Level of Service. Transportation

Engineering: Theory, Practice and Modeling, 2017, pp. 197-292.

Transportation Research Board. Highway Capacity Manual, Sixth Edition: A Guide for

Multimodal Mobility Analysis. Washington, DC, 2016.

Treiber, M., Kesting, A., and Helbing, D. Three-phase Traffic Theory and Two-phase models

with a Fundamental Diagram in the Light of Empirical Stylized Facts. Transportation

Research Part B: Methodological, Vol. 44, No. 8-9, 2010, pp. 983-1000.

Tsugawa, S. An Overview on an Automated Truck Platoon Within the Energy ITS Project.

IFAC Proceedings, Vol. 46, No. 21, 2013, pp. 41-46.

van Arem, B., Hogema, J., and Vanderschuren, M. An Assessment of the Impact of Autonomous

Intelligent Cruise Control. TNO, 1995.

van Arem, B., van Driel, C.J.G., and Visser, R. The Impact of Cooperative Adaptive Cruise

Control on Traffic-Flow Characteristics. IEEE Transactions on Intelligent


van Maarseveen, S. Impacts of Truck Platooning at Motorway On-Ramps. TuDelft, Delft,

Netherlands, 2017.

VanderWerf, J., Shladover, S., Kourjanskaia, N., Miller, M., and Krishnan, H. Modeling Effects

of Driver Control Assistance Systems on Traffic. Transportation Research Record:

Journal of the Transportation Research Board, No. 1748, 2001, pp. 167-174

VanderWerf, J., Shladover, S., Miller, M., and Kourjanskaia, N. Effects of Adaptive Cruise

Control Systems on Highway Traffic Flow Capacity. Transportation Research Record:

Journal of the Transportation Research Board, No. 1800, 2002, pp. 78-84.

Virginia Department of Motor Vehicles. Automated Vehicles in Virginia: The Policy and

Regulatory Environment. Richmond, 2017.

Virginia Department of Motor Vehicles. Virginia Driver's Manual. Richmond, 2018.

67

Virginia Department of Transportation. Traffic Operations and Safety Analysis Manual, Version

1.0. Richmond, 2015.

Virginia Department of Transportation. CY 2018-2021 Business Plan. Richmond, 2017.

Virginia Department of Transportation. Average Daily Traffic Volumes With Vehicle

Classification Data on Interstate, Arterial, and Primary Routes. Richmond, 2018a.

Virginia Department of Transportation. Daily Traffic Volume Estimates Including Vehicle

Classification Estimates. Richmond, 2018b.

Zhao, L., and Sun, J. Simulation Framework for Vehicle Platooning and Car-Following

Behaviors Under Connected-Vehicle Environment. Procedia, Social and Behavioral

Sciences, Vol. 96, 2013, pp. 914-924.

68

69

APPENDIX A

VALUES OBTAINED FROM PREVIOUS RESEARCH

70

71

Table A1. Operational Parameters From Previous Research

Source

Vehicle

Automation

(N/A, LV,

AV, CAV)

Vehicle

Type

Vehicle

Length

(ft)

Speed

(ft/s)

Comfort.

Accel.

(ft/s2)

Max.

Accel.

(ft/s2)

Comfort.

Decel.

(ft/s2)

Max.

Braking

(ft/s2)

Desired

Headway (sec)

Desired Spacing

(ft)

Min. Distance,

Standstill

Distance (ft)

L U L U L U L U L U L U L U L U

Raza and

Ioannou 1996

CAV P 88 4.83 3.22 0.416

AV P 88 4.83 3.22 0.72

Carbaugh et al.

1998

N/A P 65.62 131.23 13.12 32.81 13.12 32.81

Treiber et al.

2010

LV P 16.4 109.34 2.39 5.48 2.39 5.48 1.6

Huang et al.

2000

LV P 82.02 15

AV P 82.02 15 1.0 32.81

VanderWerf et

al. 2001

LV P

AV P 1.4

CAV P 0.5

VanderWerf et

al. 2002

AV P 95.14 9.65 9.65 1.4

CAV P 95.14 9.65 9.65 0.5

LV P 95.14 1.1

van Arem et al.

2006

LV P 6.56 8.2 1.4

CAV P 6.56 6.56 0.5

Kesting et al.

2008

N/A P 77.47 109.36 2.29 6.56 2.29 6.56

LV P 16.4 52.82 53.81 4.99 5.18 2.01 2.48 1.3 1.39 3.41 5.28

Kesting et al.

2010

LV P

77.47 109.36 2.29 6.56 2.29 6.56 1.5 2

LV P 109.36 4.59 6.56 8 1.5 6.56

LV T 77.46 2.3 6.56 2 13.12

Arnaout and

Bowling 2011

CAV P 13.12

CAV P 13.12 0.5 1

LV P 16.4 110 8.2 8.2

Shladover et al.

2012

LV P 15.42 95.33 6.56 6.56 1.48 1.8

AV P 15.42 95.33 6.56 6.56 1.1 2.2

CAV P 15.42 95.33 6.56 6.56 0.6 1.1

Fishelson et al.

2013

CAV P 98.42

19.69 39.3

7

Zhao and Sun

2013

AV P 1.4

CAV P 0.5

Desiraju et al.

2015

N/A P 16.4 98.4 0 6.56 6.56

72

Delis et al.

2015

AV P 100.25 118.48 1 2

CAV P 100.25 118.48 1 2

Ntousakis et al.

2015

LV P 109.36 0.8 2

AV P 109.36 0.8 2

AV P 119.99 4.43 5.91

Le Vine et al.

2016

LV P 45.57

AV P 26.43 4.43 4.43

AV P 10.94 1.9 1.77

Meissner et al.

2016

CAV P 16.4 85.3 111.55 9.18 10.5 9.18 10.5

Talebpour and

Mahmassani

2016

CAV P 82.02 13.12 18.02 2 6.56

CAV P 82.02 4.59 6.56 1.5 6.56

LV P 82.02 13.12 26.25

Shelton et al.

2016

LV P 85.07 105.6

CAV P 85.07 105.6

Songchitruksa

et al. 2016

CAV P 95.33 102.67 11.15 0.6 1.1

Li et al. 2017 LV P 16.4 109.36 9.18 15.748 9.18 15.748 1.6

CAV P 16.4 109.36 9.18 15.748 9.18 15.748 0.6 1.6

van

Maarseveen

2017

N/A P 13.75 118.47 4.1 19.69 1.2

N/A T

(Light)

27.89 77.46 2.62 14.76 1.2

N/A T

(Heavy)

54.13 77.46 1.31 13.12 1.5

CAV T

(Heavy)

54.13 77.46 1.31 13.12 1.5

Melson et al.

2018

LV P 14.6 73.33 6.5

CAV P 14.6 73.33 0.6

N/A = not applicable; LV = legacy vehicles; AV = automated vehicles; CAV = connected automated vehicles; P = passenger vehicles; T = trucks; Comfort. = comfortable; Accel. = acceleration; decel

= deceleration; U = upper value; L = lower value.

73

APPENDIX B

HEAVY VEHICLE MARKET PENETRATION SCENARIOS

Heavy Vehicles on the Test Network

Scenarios 200 to 220 represent 5% HV, scenarios 300 to 320 represent 10% HV, and

scenarios 400 to 420 represent 15% HV.

Table B1. Market Penetration of Heavy Vehicle Scenarios

Scenario No.

Percentage of Passenger

Vehicles

Percentage of Light Trucks

Percentage of Heavy Trucks

LV AV CAV LV AV CAV LV AV CAV

200 95.00 0.00 0.00 0.90 0.00 0.00 4.10 0.00 0.00

201 76.00 19.00 0.00 0.72 0.18 0.00 3.28 0.82 0.00

202 76.00 0.00 19.00 0.72 0.00 0.18 3.28 0.00 0.82

203 57.00 38.00 0.00 0.54 0.36 0.00 2.46 1.64 0.00

204 57.00 19.00 19.00 0.54 0.18 0.18 2.46 0.82 0.82

205 57.00 0.00 38.00 0.54 0.00 0.36 2.46 0.00 1.64

206 38.00 57.00 0.00 0.36 0.54 0.00 1.64 2.46 0.00

207 38.00 38.00 19.00 0.36 0.36 0.18 1.64 1.64 0.82

208 38.00 19.00 38.00 0.36 0.18 0.36 1.64 0.82 1.64

209 38.00 0.00 57.00 0.36 0.00 0.54 1.64 0.00 2.46

210 19.00 76.00 0.00 0.18 0.72 0.00 0.82 3.28 0.00

211 19.00 57.00 19.00 0.18 0.54 0.18 0.82 2.46 0.82

212 19.00 38.00 38.00 0.18 0.36 0.36 0.82 1.64 1.64

213 19.00 19.00 57.00 0.18 0.18 0.54 0.82 0.82 2.46

214 19.00 0.00 76.00 0.18 0.00 0.72 0.82 0.00 3.28

215 0.00 95.00 0.00 0.00 0.90 0.00 0.00 4.10 0.00

216 0.00 76.00 19.00 0.00 0.72 0.18 0.00 3.28 0.82

217 0.00 57.00 38.00 0.00 0.54 0.36 0.00 2.46 1.64

218 0.00 38.00 57.00 0.00 0.36 0.54 0.00 1.64 2.46

219 0.00 19.00 76.00 0.00 0.18 0.72 0.00 0.82 3.28

220 0.00 0.00 95.00 0.00 0.00 0.90 0.00 0.00 4.10

300 90.00 0.00 0.00 1.80 0.00 0.00 8.20 0.00 0.00

301 72.00 18.00 0.00 1.44 0.36 0.00 6.56 1.64 0.00

302 72.00 0.00 18.00 1.44 0.00 0.36 6.56 0.00 1.64

303 54.00 36.00 0.00 1.08 0.72 0.00 4.92 3.28 0.00

304 54.00 18.00 18.00 1.08 0.36 0.36 4.92 1.64 1.64

305 54.00 0.00 36.00 1.08 0.00 0.72 4.92 0.00 3.28

306 36.00 54.00 0.00 0.72 1.08 0.00 3.28 4.92 0.00

307 36.00 36.00 18.00 0.72 0.72 0.36 3.28 3.28 1.64

308 36.00 18.00 36.00 0.72 0.36 0.72 3.28 1.64 3.28

309 36.00 0.00 54.00 0.72 0.00 1.08 3.28 0.00 4.92

310 18.00 72.00 0.00 0.36 1.44 0.00 1.64 6.56 0.00

311 18.00 54.00 18.00 0.36 1.08 0.36 1.64 4.92 1.64

312 18.00 36.00 36.00 0.36 0.72 0.72 1.64 3.28 3.28

313 18.00 18.00 54.00 0.36 0.36 1.08 1.64 1.64 4.92

314 18.00 0.00 72.00 0.36 0.00 1.44 1.64 0.00 6.56

315 0.00 90.00 0.00 0.00 1.80 0.00 0.00 8.20 0.00

316 0.00 72.00 18.00 0.00 1.44 0.36 0.00 6.56 1.64

317 0.00 54.00 36.00 0.00 1.08 0.72 0.00 4.92 3.28

74

Scenario No.


Vehicles


Percentage of Heavy Trucks


318 0.00 36.00 54.00 0.00 0.72 1.08 0.00 3.28 4.92

319 0.00 18.00 72.00 0.00 0.36 1.44 0.00 1.64 6.56

320 0.00 0.00 90.00 0.00 0.00 1.80 0.00 0.00 8.20

400 85.00 0.00 0.00 2.70 0.00 0.00 12.30 0.00 0.00

401 68.00 17.00 0.00 2.16 0.54 0.00 9.84 2.46 0.00

402 68.00 0.00 17.00 2.16 0.00 0.54 9.84 0.00 2.46

403 51.00 34.00 0.00 1.62 1.08 0.00 7.38 4.92 0.00

404 51.00 17.00 17.00 1.62 0.54 0.54 7.38 2.46 2.46

405 51.00 0.00 34.00 1.62 0.00 1.08 7.38 0.00 4.92

406 34.00 51.00 0.00 1.08 1.62 0.00 4.92 7.38 0.00

407 34.00 34.00 17.00 1.08 1.08 0.54 4.92 4.92 2.46

408 34.00 17.00 34.00 1.08 0.54 1.08 4.92 2.46 4.92

409 34.00 0.00 51.00 1.08 0.00 1.62 4.92 0.00 7.38

410 17.00 68.00 0.00 0.54 2.16 0.00 2.46 9.84 0.00

411 17.00 51.00 17.00 0.54 1.62 0.54 2.46 7.38 2.46

412 17.00 34.00 34.00 0.54 1.08 1.08 2.46 4.92 4.92

413 17.00 17.00 51.00 0.54 0.54 1.62 2.46 2.46 7.38

414 17.00 0.00 68.00 0.54 0.00 2.16 2.46 0.00 9.84

415 0.00 85.00 0.00 0.00 2.70 0.00 0.00 12.30 0.00

416 0.00 68.00 17.00 0.00 2.16 0.54 0.00 9.84 2.46

417 0.00 51.00 34.00 0.00 1.62 1.08 0.00 7.38 4.92

418 0.00 34.00 51.00 0.00 1.08 1.62 0.00 4.92 7.38

419 0.00 17.00 68.00 0.00 0.54 2.16 0.00 2.46 9.84

420 0.00 0.00 85.00 0.00 0.00 2.70 0.00 0.00 12.30


HVs on the I-81 Corridor

Scenarios 100 to 120 represent passenger vehicles only, scenarios 300 to 320 represent

30% HV, and scenarios 500 to 520 represent 50% HV.

Table B2. Market Penetration of Heavy Vehicle Scenarios

Scenario No.


Vehicles (%)


(%)

Percentage Heavy Trucks

(%)


100 100.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

101 80.00 20.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

102 80.00 0.00 20.00 0.00 0.00 0.00 0.00 0.00 0.00

103 60.00 40.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

104 60.00 20.00 20.00 0.00 0.00 0.00 0.00 0.00 0.00

105 60.00 0.00 40.00 0.00 0.00 0.00 0.00 0.00 0.00

106 40.00 60.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

107 40.00 40.00 20.00 0.00 0.00 0.00 0.00 0.00 0.00

108 40.00 20.00 40.00 0.00 0.00 0.00 0.00 0.00 0.00

109 40.00 0.00 60.00 0.00 0.00 0.00 0.00 0.00 0.00

75

Scenario No.


Vehicles (%)


(%)


(%)


110 20.00 80.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

111 20.00 60.00 20.00 0.00 0.00 0.00 0.00 0.00 0.00

112 20.00 40.00 40.00 0.00 0.00 0.00 0.00 0.00 0.00

113 20.00 20.00 60.00 0.00 0.00 0.00 0.00 0.00 0.00

114 20.00 0.00 80.00 0.00 0.00 0.00 0.00 0.00 0.00

115 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

116 0.00 80.00 20.00 0.00 0.00 0.00 0.00 0.00 0.00

117 0.00 60.00 40.00 0.00 0.00 0.00 0.00 0.00 0.00

118 0.00 40.00 60.00 0.00 0.00 0.00 0.00 0.00 0.00

119 0.00 20.00 80.00 0.00 0.00 0.00 0.00 0.00 0.00

120 0.00 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0.00

300 70.00 0.00 0.00 10.50 0.00 0.00 19.50 0.00 0.00

301 56.00 14.00 0.00 8.40 2.10 0.00 15.60 3.90 0.00

302 56.00 0.00 14.00 8.40 0.00 2.10 15.60 0.00 3.90

303 42.00 28.00 0.00 6.30 4.20 0.00 11.70 7.80 0.00

304 42.00 14.00 14.00 6.30 2.10 2.10 11.70 3.90 3.90

305 42.00 0.00 28.00 6.30 0.00 4.20 11.70 0.00 7.80

306 28.00 42.00 0.00 4.20 6.30 0.00 7.80 11.70 0.00

307 28.00 28.00 14.00 4.20 4.20 2.10 7.80 7.80 3.90

308 28.00 14.00 28.00 4.20 2.10 4.20 7.80 3.90 7.80

309 28.00 0.00 42.00 4.20 0.00 6.30 7.80 0.00 11.70

310 14.00 56.00 0.00 2.10 8.40 0.00 3.90 15.60 0.00

311 14.00 42.00 14.00 2.10 6.30 2.10 3.90 11.70 3.90

312 14.00 28.00 28.00 2.10 4.20 4.20 3.90 7.80 7.80

313 14.00 14.00 42.00 2.10 2.10 6.30 3.90 3.90 11.70

314 14.00 0.00 56.00 2.10 0.00 8.40 3.90 0.00 15.60

315 0.00 70.00 0.00 0.00 10.50 0.00 0.00 19.50 0.00

316 0.00 56.00 14.00 0.00 8.40 2.10 0.00 15.60 3.90

317 0.00 42.00 28.00 0.00 6.30 4.20 0.00 11.70 7.80

318 0.00 28.00 42.00 0.00 4.20 6.30 0.00 7.80 11.70

319 0.00 14.00 56.00 0.00 2.10 8.40 0.00 3.90 15.60

320 0.00 0.00 70.00 0.00 0.00 10.50 0.00 0.00 19.50

500 50.00 0.00 0.00 17.50 0.00 0.00 32.50 0.00 0.00

501 40.00 10.00 0.00 14.00 3.50 0.00 26.00 6.50 0.00

502 40.00 0.00 10.00 14.00 0.00 3.50 26.00 0.00 6.50

503 30.00 20.00 0.00 10.50 7.00 0.00 19.50 13.00 0.00

504 30.00 10.00 10.00 10.50 3.50 3.50 19.50 6.50 6.50

505 30.00 0.00 20.00 10.50 0.00 7.00 19.50 0.00 13.00

506 20.00 30.00 0.00 7.00 10.50 0.00 13.00 19.50 0.00

507 20.00 20.00 10.00 7.00 7.00 3.50 13.00 13.00 6.50

508 20.00 10.00 20.00 7.00 3.50 7.00 13.00 6.50 13.00

509 20.00 0.00 30.00 7.00 0.00 10.50 13.00 0.00 19.50

510 10.00 40.00 0.00 3.50 14.00 0.00 6.50 26.00 0.00

511 10.00 30.00 10.00 3.50 10.50 3.50 6.50 19.50 6.50

512 10.00 20.00 20.00 3.50 7.00 7.00 6.50 13.00 13.00

513 10.00 10.00 30.00 3.50 3.50 10.50 6.50 6.50 19.50

514 10.00 0.00 40.00 3.50 0.00 14.00 6.50 0.00 26.00

76

Scenario No.


Vehicles (%)


(%)


(%)


515 0.00 50.00 0.00 0.00 17.50 0.00 0.00 32.50 0.00

516 0.00 40.00 10.00 0.00 14.00 3.50 0.00 26.00 6.50

517 0.00 30.00 20.00 0.00 10.50 7.00 0.00 19.50 13.00

518 0.00 20.00 30.00 0.00 7.00 10.50 0.00 13.00 19.50

519 0.00 10.00 40.00 0.00 3.50 14.00 0.00 6.50 26.00

520 0.00 0.00 50.00 0.00 0.00 17.50 0.00 0.00 32.50


77

APPENDIX C

STATISTICAL RESULTS

T-Tests

P-value results from the paired two-tailed t-test were obtained from Excel. P-values

greater than 0.1 indicate an insignificant difference between compared samples. The results for

each case study are summarized here. Only results indicating insignificant difference p-value >

0.1 were reported in this case; all other (unreported) outputs indicated p-value < 0.1.

I-95

Table C1. T-Test From Throughput for 100% Demand Northbound

Northbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

Willis Rd. and SR-288 0% 9.21E-01* -2% 2.48E-05

SR-288 0% 9.73E-01* -2% 2.26E-06

SR-288 and SR-10 0% 3.48E-01* -1% 9.84E-03

* P-value > 0.1.

Table C2. T-Test From Throughput for 150% Demand Northbound

Northbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

Maury St. and I-195 0% 2.7E-01* -10% 2.6E-04

Belles Rd 1% 1.7E-01* -5% 4.0E-05

Belles Rd and SR-150 0% 4.0E-01* -6% 2.5E-07

* P-value > 0.1.

Table C3. T-Test from Throughput for 200% Demand Northbound

Northbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

Maury St. and I-195 7% 3.41E-06 1% 3.73E-01*

* P-value > 0.1.

Table C4. T-Test From Average Speed for 100% Demand Southbound

Southbound


AV and LV

P-Value


CAV and LV

P-Value

SR-150 and Willis Rd. 3% 5.2E-05 0% 8.7E-01*

Willis Rd. and SR-288 3% 1.5E-05 -1% 3.6E-01*

SR-288 and SR-10 4% 5.1E-06 -1% 4.9E-01*

SR-10 3% 2.4E-06 1% 2.3E-01*

* P-value > 0.1.

78

Table C5. T-Test From Average Speed for 100% Demand Northbound

Northbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

SR-150 and Willis Rd. -1% 1.4E-01* -4% 2.4E-03

SR-288 and SR-10 3% 4.7E-07 0% 4.2E-01*

SR-10 0% 2.6E-01* -1% 2.5E-04

* P-value > 0.1.

Table C6. T-Test from Average Speed for 150% Demand Southbound

Southbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

Willis Rd. 6% 5.62E-08 -2% 2.00E-01*

Willis Rd. and SR-288 9% 8.58E-08 2% 2.12E-01*

SR-288 7% 1.72E-05 -1% 5.22E-01*

SR-288 and SR-10 8% 8.15E-07 1% 6.05E-01*

SR-10 2% 8.59E-05 0% 9.76E-01*

* P-value > 0.1.

Table C7. T-Test from Average Speed for 150% Demand Northbound

Northbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

SR-150 and Willis Rd. 5% 2.13E-05 1% 5.14E-01*

Willis Rd. 3% 5.79E-03 -1% 5.33E-01*


* P-value > 0.1.

Table C8. T-Test from Average Speed for 200% Demand Northbound

Northbound

% Difference

Between AV and LV

P-Value


CAV and LV

P-Value

Maury St. and Belles Rd. 1% 1.29E-01* -5% 6.01E-04

Belles Rd 0% 6.07E-01* -7% 2.53E-04

Willis Rd. 4% 4.54E-03 -1% 5.27E-01*


* P-value > 0.1.

I-81

Scenarios varied by vehicle composition with HV percentages of 0%, 30%, and 50%. In

each of the three HV composition cases, MPs of AVs and CAVs were increased by intervals of

20 percentage points, resulting in 21 MP scenarios per case (LV, AV, CAV combinations, as

shown in Table B2). Scenarios were numbered 100 to 120 (for 0% HV), 300 to 320 (for 30%

HV), and 500 to 520 (for 50% HV).

79

Table C9. T-Test From Speed Across Scenarios

Scenario Speed (mph) % Difference From LV Only P-Value

100 63.99 0% 101 64.10 0% 2.57E-01*

102 64.08 0% 3.85E-01*

103 64.23 0% 1.67E-01*

105 64.37 1% 1.13E-01*

303 39.01 5% 1.48E-01*

306 38.56 4% 3.06E-01*

501 36.65 2% 6.29E-01*

502 38.39 7% 1.43E-01*

* P-value > 0.1.

Table C10. T-Test from Speed Across Heavy Vehicle Scenarios

Compared Scenarios % Difference Between HV% P-Value

300 vs. 500 3% 3.8E-01*

301 vs. 501 8% 1.2E-01*

303 vs. 503 -1% 8.0E-01*

305 vs. 505 0% 9.3E-01*

306 vs. 506 -1% 7.6E-01*

307 vs. 507 -1% 6.1E-01*

309 vs. 509 5% 1.2E-01*

310 vs. 510 -2% 4.8E-01*

311 vs. 511 5% 1.2E-01*

315 vs. 515 1% 3.9E-01*

316 vs. 516 3% 2.3E-01*

317 vs. 517 2% 4.3E-01*

318 vs. 518 4% 1.5E-01*

320 vs. 520 5% 2.5E-01*

* P-value > 0.1.

80

Speed Results

Table C11. Speed Comparison Across Heavy Vehicle Scenarios

Compared Scenarios


0% and 30% HV

Compared Scenarios


30% and 50% HV

100 vs. 300 42% 300 vs. 500 3% 101 vs. 301 38% 301 vs. 501 8%

102 vs. 302 35% 302 vs. 502 8%

103 vs. 303 39% 303 vs. 503 -1%

104 vs. 304 31% 304 vs. 504 9%

105 vs. 305 34% 305 vs. 505 0%

106 vs. 306 40% 306 vs. 506 -1%

107 vs. 307 35% 307 vs. 507 -1%

108 vs. 308 30% 308 vs. 508 7%

109 vs. 309 28% 309 vs. 509 5%

110 vs. 310 39% 310 vs. 510 -2%

111 vs. 311 33% 311 vs. 511 5%

112 vs. 312 31% 312 vs. 512 5%

113 vs. 313 27% 313 vs. 513 5%

114 vs. 314 19% 314 vs. 514 8%

115 vs. 315 40% 315 vs. 515 1% 116 vs. 316 36% 316 vs. 516 3% 117 vs. 317 32% 317 vs. 517 2% 118 vs. 318 30% 318 vs. 518 4% 119 vs. 319 25% 319 vs. 519 6% 120 vs. 320 21% 320 vs. 520 5%

Documents

Assessment of Capacity Changes Due to Automated ...Standard Title Page - Report on Federally Funded Project 1. Report No.: 2. Government Accession No.: 3. Recipient’s Catalog No.: